Upload folder using huggingface_hub

README.md

@@ -0,0 +1,285 @@
+---
+license: apache-2.0
+language:
+  - en
+tags:
+  - image-quality-assessment
+  - document-quality
+  - mplug-owl2
+  - vision-language
+  - document-analysis
+  - sharpness
+  - blur-detection
+  - IQA
+pipeline_tag: image-to-text
+library_name: transformers
+---
+
+# DeQA-Doc-Sharpness: Document Image Sharpness Assessment
+
+**DeQA-Doc-Sharpness** is a vision-language model specialized in assessing the **sharpness and clarity** of document images. It evaluates focus quality, blur levels, and text legibility in scanned or photographed documents.
+
+## Model Family
+
+This model is part of the **DeQA-Doc** family, which includes three specialized models:
+
+| Model | Description | HuggingFace |
+|-------|-------------|-------------|
+| **DeQA-Doc-Overall** | Overall document quality | [mapo80/DeQA-Doc-Overall](https://huggingface.co/mapo80/DeQA-Doc-Overall) |
+| **DeQA-Doc-Color** | Color quality assessment | [mapo80/DeQA-Doc-Color](https://huggingface.co/mapo80/DeQA-Doc-Color) |
+| **DeQA-Doc-Sharpness** | Sharpness/clarity assessment (this model) | [mapo80/DeQA-Doc-Sharpness](https://huggingface.co/mapo80/DeQA-Doc-Sharpness) |
+
+## Quick Start
+
+```python
+import torch
+from transformers import AutoModelForCausalLM
+from PIL import Image
+
+# Load the model
+model = AutoModelForCausalLM.from_pretrained(
+    "mapo80/DeQA-Doc-Sharpness",
+    trust_remote_code=True,
+    torch_dtype=torch.float16,
+    device_map="auto",
+)
+
+# Score an image
+image = Image.open("document.jpg").convert("RGB")
+score = model.score([image])
+print(f"Sharpness Score: {score.item():.2f} / 5.0")
+```
+
+## What Does Sharpness Quality Measure?
+
+The sharpness score evaluates:
+
+- **Focus Quality**: How well the document is in focus
+- **Motion Blur**: Absence of blur from camera/scanner movement
+- **Text Clarity**: Sharpness of text edges and characters
+- **Detail Preservation**: Fine details are visible and crisp
+- **Resolution Quality**: Adequate resolution for the content
+
+## Score Interpretation
+
+| Score Range | Quality Level | Typical Issues |
+|-------------|---------------|----------------|
+| 4.5 - 5.0 | **Excellent** | Perfectly sharp, crisp text |
+| 3.5 - 4.5 | **Good** | Slight softness, still very readable |
+| 2.5 - 3.5 | **Fair** | Noticeable blur, readable with effort |
+| 1.5 - 2.5 | **Poor** | Significant blur, hard to read |
+| 1.0 - 1.5 | **Bad** | Severe blur, text illegible |
+
+## Batch Processing
+
+```python
+images = [
+    Image.open("doc1.jpg").convert("RGB"),
+    Image.open("doc2.jpg").convert("RGB"),
+    Image.open("doc3.jpg").convert("RGB"),
+]
+
+scores = model.score(images)
+for i, score in enumerate(scores):
+    print(f"Document {i+1} Sharpness: {score.item():.2f} / 5.0")
+```
+
+## Use Cases
+
+- **OCR Preprocessing**: Filter blurry images before OCR to improve accuracy
+- **Document Capture QA**: Real-time feedback for mobile document scanning
+- **Archive Quality Control**: Identify documents needing re-scanning
+- **Blur Detection**: Automatic detection of out-of-focus captures
+- **Scanner Maintenance**: Detect scanner focus issues
+
+## Example: OCR Quality Gate
+
+```python
+import torch
+from transformers import AutoModelForCausalLM
+from PIL import Image
+
+model = AutoModelForCausalLM.from_pretrained(
+    "mapo80/DeQA-Doc-Sharpness",
+    trust_remote_code=True,
+    torch_dtype=torch.float16,
+    device_map="auto",
+)
+
+def check_ocr_readiness(image_path, min_sharpness=3.5):
+    """Check if document is sharp enough for reliable OCR."""
+    img = Image.open(image_path).convert("RGB")
+    score = model.score([img]).item()
+
+    if score >= min_sharpness:
+        return True, score, "Ready for OCR"
+    elif score >= 2.5:
+        return False, score, "May produce OCR errors - consider rescanning"
+    else:
+        return False, score, "Too blurry for OCR - rescan required"
+
+ready, score, message = check_ocr_readiness("scan.jpg")
+print(f"Sharpness: {score:.2f}/5.0 - {message}")
+
+if ready:
+    # Proceed with OCR
+    pass
+else:
+    # Request rescan
+    pass
+```
+
+## Example: Batch Quality Sorting
+
+```python
+import torch
+from transformers import AutoModelForCausalLM
+from PIL import Image
+from pathlib import Path
+
+model = AutoModelForCausalLM.from_pretrained(
+    "mapo80/DeQA-Doc-Sharpness",
+    trust_remote_code=True,
+    torch_dtype=torch.float16,
+    device_map="auto",
+)
+
+def sort_by_sharpness(image_folder):
+    """Sort documents into quality buckets based on sharpness."""
+    results = {"excellent": [], "good": [], "fair": [], "poor": [], "bad": []}
+
+    for img_path in Path(image_folder).glob("*.jpg"):
+        img = Image.open(img_path).convert("RGB")
+        score = model.score([img]).item()
+
+        if score >= 4.5:
+            results["excellent"].append((img_path, score))
+        elif score >= 3.5:
+            results["good"].append((img_path, score))
+        elif score >= 2.5:
+            results["fair"].append((img_path, score))
+        elif score >= 1.5:
+            results["poor"].append((img_path, score))
+        else:
+            results["bad"].append((img_path, score))
+
+    return results
+
+# Usage
+quality_report = sort_by_sharpness("scanned_docs/")
+print(f"Excellent: {len(quality_report['excellent'])} documents")
+print(f"Need rescan: {len(quality_report['poor']) + len(quality_report['bad'])} documents")
+```
+
+## Multi-Dimensional Quality Assessment
+
+Combine with other DeQA-Doc models for comprehensive assessment:
+
+```python
+import torch
+from transformers import AutoModelForCausalLM
+from PIL import Image
+
+# Load all three models
+models = {
+    "overall": AutoModelForCausalLM.from_pretrained(
+        "mapo80/DeQA-Doc-Overall", trust_remote_code=True,
+        torch_dtype=torch.float16, device_map="auto"
+    ),
+    "color": AutoModelForCausalLM.from_pretrained(
+        "mapo80/DeQA-Doc-Color", trust_remote_code=True,
+        torch_dtype=torch.float16, device_map="auto"
+    ),
+    "sharpness": AutoModelForCausalLM.from_pretrained(
+        "mapo80/DeQA-Doc-Sharpness", trust_remote_code=True,
+        torch_dtype=torch.float16, device_map="auto"
+    ),
+}
+
+def full_quality_report(image_path):
+    img = Image.open(image_path).convert("RGB")
+
+    scores = {}
+    for name, model in models.items():
+        scores[name] = model.score([img]).item()
+
+    return scores
+
+report = full_quality_report("document.jpg")
+print(f"Overall:   {report['overall']:.2f}/5.0")
+print(f"Color:     {report['color']:.2f}/5.0")
+print(f"Sharpness: {report['sharpness']:.2f}/5.0")
+```
+
+## Model Architecture
+
+- **Base Model**: mPLUG-Owl2 (LLaMA2-7B + ViT-L Vision Encoder)
+- **Vision Encoder**: CLIP ViT-L/14 (1024 visual tokens via Visual Abstractor)
+- **Language Model**: LLaMA2-7B
+- **Training**: Full fine-tuning on document sharpness quality datasets
+- **Input Resolution**: Images are resized to 448x448 (with aspect ratio preservation)
+
+## Technical Details
+
+| Property | Value |
+|----------|-------|
+| Model Size | ~16 GB (float16) |
+| Parameters | ~7.2B |
+| Input | RGB images (any resolution) |
+| Output | Sharpness quality score (1.0 - 5.0) |
+| Inference | ~2-3 seconds per image on A100 |
+
+## Hardware Requirements
+
+| Setup | VRAM Required | Recommended |
+|-------|---------------|-------------|
+| Full precision (fp32) | ~32 GB | A100, H100 |
+| Half precision (fp16) | ~16 GB | A100, A40, RTX 4090 |
+| With CPU offload | ~8 GB GPU + RAM | RTX 3090, RTX 4080 |
+
+## Installation
+
+```bash
+pip install torch transformers accelerate pillow sentencepiece protobuf
+```
+
+**Note**: Use `transformers>=4.36.0` for best compatibility.
+
+## Comparison with Traditional Methods
+
+| Method | Pros | Cons |
+|--------|------|------|
+| **Laplacian Variance** | Fast, simple | Only measures edge intensity |
+| **FFT-based** | Frequency analysis | Sensitive to image content |
+| **Gradient-based** | Good for text | Requires tuning |
+| **DeQA-Doc-Sharpness** | Content-aware, trained on documents | Requires GPU |
+
+DeQA-Doc-Sharpness understands document context and can differentiate between intentionally smooth backgrounds and unintentional blur.
+
+## Limitations
+
+- Optimized for document images (text, forms, letters)
+- May not generalize well to natural photos
+- Requires GPU with sufficient VRAM for efficient inference
+- Sharpness assessment is relative to training data distribution
+
+## Citation
+
+```bibtex
+@misc{deqa-doc-sharpness-2024,
+  title={DeQA-Doc-Sharpness: Document Image Sharpness Assessment},
+  author={mapo80},
+  year={2024},
+  publisher={HuggingFace},
+  url={https://huggingface.co/mapo80/DeQA-Doc-Sharpness}
+}
+```
+
+## License
+
+Apache 2.0
+
+## Related Models
+
+- [DeQA-Doc-Overall](https://huggingface.co/mapo80/DeQA-Doc-Overall) - Overall quality assessment
+- [DeQA-Doc-Color](https://huggingface.co/mapo80/DeQA-Doc-Color) - Color quality assessment

__init__.py

@@ -0,0 +1,5 @@
+from .configuration_mplug_owl2 import MPLUGOwl2Config
+from .modeling_attn_mask_utils import _prepare_4d_causal_attention_mask
+from .modeling_mplug_owl2_huggingface import MPLUGOwl2LlamaForCausalLM
+
+__all__ = ["MPLUGOwl2Config", "MPLUGOwl2LlamaForCausalLM"]

config.json

@@ -0,0 +1,217 @@
+{
+  "_name_or_path": "./",
+  "architectures": [
+    "MPLUGOwl2LlamaForCausalLM"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_mplug_owl2.MPLUGOwl2Config",
+    "AutoModel": "modeling_mplug_owl2_huggingface.MPLUGOwl2LlamaForCausalLM",
+    "AutoModelForCausalLM": "modeling_mplug_owl2_huggingface.MPLUGOwl2LlamaForCausalLM"
+  },
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "binary_rating_loss": "fidelity",
+  "bos_token_id": 1,
+  "closeset_rating_loss": true,
+  "continuous_rating_loss": true,
+  "detach_pred_std": true,
+  "eos_token_id": 2,
+  "freeze_vision_model": false,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "image_aspect_ratio": "pad",
+  "image_grid_pinpoints": null,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "level_ids": [
+    15129,
+    1781,
+    6534,
+    6460,
+    4319
+  ],
+  "level_prefix": [
+    11029,
+    310,
+    278,
+    1967,
+    338
+  ],
+  "max_position_embeddings": 2048,
+  "model_type": "mplug_owl2",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "softkl_loss": true,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.36.1",
+  "tune_visual_abstractor": true,
+  "use_cache": true,
+  "use_fix_std": true,
+  "visual_abstractor_lr": null,
+  "visual_config": {
+    "visual_abstractor": {
+      "_name_or_path": "",
+      "add_cross_attention": false,
+      "architectures": null,
+      "attention_probs_dropout_prob": 0.0,
+      "bad_words_ids": null,
+      "begin_suppress_tokens": null,
+      "bos_token_id": null,
+      "chunk_size_feed_forward": 0,
+      "cross_attention_hidden_size": null,
+      "decoder_start_token_id": null,
+      "diversity_penalty": 0.0,
+      "do_sample": false,
+      "early_stopping": false,
+      "encoder_hidden_size": 1024,
+      "encoder_no_repeat_ngram_size": 0,
+      "eos_token_id": null,
+      "exponential_decay_length_penalty": null,
+      "finetuning_task": null,
+      "forced_bos_token_id": null,
+      "forced_eos_token_id": null,
+      "grid_size": 32,
+      "hidden_size": 1024,
+      "id2label": {
+        "0": "LABEL_0",
+        "1": "LABEL_1"
+      },
+      "initializer_range": 0.02,
+      "intermediate_size": 2816,
+      "is_decoder": false,
+      "is_encoder_decoder": false,
+      "label2id": {
+        "LABEL_0": 0,
+        "LABEL_1": 1
+      },
+      "layer_norm_eps": 1e-06,
+      "length_penalty": 1.0,
+      "max_length": 20,
+      "min_length": 0,
+      "model_type": "mplug_owl_visual_abstract",
+      "no_repeat_ngram_size": 0,
+      "num_attention_heads": 16,
+      "num_beam_groups": 1,
+      "num_beams": 1,
+      "num_hidden_layers": 6,
+      "num_learnable_queries": 64,
+      "num_return_sequences": 1,
+      "output_attentions": false,
+      "output_hidden_states": false,
+      "output_scores": false,
+      "pad_token_id": null,
+      "prefix": null,
+      "problem_type": null,
+      "pruned_heads": {},
+      "remove_invalid_values": false,
+      "repetition_penalty": 1.0,
+      "return_dict": true,
+      "return_dict_in_generate": false,
+      "sep_token_id": null,
+      "suppress_tokens": null,
+      "task_specific_params": null,
+      "temperature": 1.0,
+      "tf_legacy_loss": false,
+      "tie_encoder_decoder": false,
+      "tie_word_embeddings": true,
+      "tokenizer_class": null,
+      "top_k": 50,
+      "top_p": 1.0,
+      "torch_dtype": null,
+      "torchscript": false,
+      "transformers_version": "4.28.1",
+      "typical_p": 1.0,
+      "use_bfloat16": false
+    },
+    "visual_model": {
+      "_name_or_path": "",
+      "add_cross_attention": false,
+      "architectures": null,
+      "attention_dropout": 0.0,
+      "bad_words_ids": null,
+      "begin_suppress_tokens": null,
+      "bos_token_id": null,
+      "chunk_size_feed_forward": 0,
+      "cross_attention_hidden_size": null,
+      "decoder_start_token_id": null,
+      "diversity_penalty": 0.0,
+      "do_sample": false,
+      "early_stopping": false,
+      "encoder_no_repeat_ngram_size": 0,
+      "eos_token_id": null,
+      "exponential_decay_length_penalty": null,
+      "finetuning_task": null,
+      "forced_bos_token_id": null,
+      "forced_eos_token_id": null,
+      "hidden_act": "quick_gelu",
+      "hidden_size": 1024,
+      "id2label": {
+        "0": "LABEL_0",
+        "1": "LABEL_1"
+      },
+      "image_size": 448,
+      "initializer_factor": 1.0,
+      "initializer_range": 0.02,
+      "intermediate_size": 4096,
+      "is_decoder": false,
+      "is_encoder_decoder": false,
+      "label2id": {
+        "LABEL_0": 0,
+        "LABEL_1": 1
+      },
+      "layer_norm_eps": 1e-06,
+      "length_penalty": 1.0,
+      "max_length": 20,
+      "min_length": 0,
+      "model_type": "mplug_owl_vision_model",
+      "no_repeat_ngram_size": 0,
+      "num_attention_heads": 16,
+      "num_beam_groups": 1,
+      "num_beams": 1,
+      "num_channels": 3,
+      "num_hidden_layers": 24,
+      "num_return_sequences": 1,
+      "output_attentions": false,
+      "output_hidden_states": false,
+      "output_scores": false,
+      "pad_token_id": null,
+      "patch_size": 14,
+      "prefix": null,
+      "problem_type": null,
+      "projection_dim": 768,
+      "pruned_heads": {},
+      "remove_invalid_values": false,
+      "repetition_penalty": 1.0,
+      "return_dict": true,
+      "return_dict_in_generate": false,
+      "sep_token_id": null,
+      "suppress_tokens": null,
+      "task_specific_params": null,
+      "temperature": 1.0,
+      "tf_legacy_loss": false,
+      "tie_encoder_decoder": false,
+      "tie_word_embeddings": true,
+      "tokenizer_class": null,
+      "top_k": 50,
+      "top_p": 1.0,
+      "torch_dtype": null,
+      "torchscript": false,
+      "transformers_version": "4.28.1",
+      "typical_p": 1.0,
+      "use_bfloat16": false,
+      "use_flash_attn": false
+    }
+  },
+  "vocab_size": 32000,
+  "weight_desp": 1.0,
+  "weight_in_level": null,
+  "weight_next_token": 0.01,
+  "weight_rank": 1.0,
+  "weight_softkl": 1.0
+}

configuration_mplug_owl2.py

@@ -0,0 +1,334 @@
+# Copyright (c) Alibaba.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+import copy
+import os
+from typing import Union
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.models.auto.modeling_auto import MODEL_FOR_CAUSAL_LM_MAPPING_NAMES
+from transformers.utils import logging
+from transformers.models.auto import CONFIG_MAPPING
+
+
+class LlamaConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`LlamaModel`]. It is used to instantiate an LLaMA
+    model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
+    defaults will yield a similar configuration to that of the LLaMA-7B.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 32000):
+            Vocabulary size of the LLaMA model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`LlamaModel`]
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 11008):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer decoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer decoder.
+        num_key_value_heads (`int`, *optional*):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details checkout [this
+            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
+            `num_attention_heads`.
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        max_position_embeddings (`int`, *optional*, defaults to 2048):
+            The maximum sequence length that this model might ever be used with. Llama 1 supports up to 2048 tokens,
+            Llama 2 up to 4096, CodeLlama up to 16384.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-06):
+            The epsilon used by the rms normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        pad_token_id (`int`, *optional*):
+            Padding token id.
+        bos_token_id (`int`, *optional*, defaults to 1):
+            Beginning of stream token id.
+        eos_token_id (`int`, *optional*, defaults to 2):
+            End of stream token id.
+        pretraining_tp (`int`, *optional*, defaults to 1):
+            Experimental feature. Tensor parallelism rank used during pretraining. Please refer to [this
+            document](https://huggingface.co/docs/transformers/parallelism) to understand more about it. This value is
+            necessary to ensure exact reproducibility of the pretraining results. Please refer to [this
+            issue](https://github.com/pytorch/pytorch/issues/76232).
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether to tie weight embeddings
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the RoPE embeddings.
+        rope_scaling (`Dict`, *optional*):
+            Dictionary containing the scaling configuration for the RoPE embeddings. Currently supports two scaling
+            strategies: linear and dynamic. Their scaling factor must be a float greater than 1. The expected format is
+            `{"type": strategy name, "factor": scaling factor}`. When using this flag, don't update
+            `max_position_embeddings` to the expected new maximum. See the following thread for more information on how
+            these scaling strategies behave:
+            https://www.reddit.com/r/LocalLLaMA/comments/14mrgpr/dynamically_scaled_rope_further_increases/. This is an
+            experimental feature, subject to breaking API changes in future versions.
+        attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`):
+            Whether to use a bias in the query, key, value and output projection layers during self-attention.
+
+
+    ```python
+    >>> from transformers import LlamaModel, LlamaConfig
+
+    >>> # Initializing a LLaMA llama-7b style configuration
+    >>> configuration = LlamaConfig()
+
+    >>> # Initializing a model from the llama-7b style configuration
+    >>> model = LlamaModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+    model_type = "llama"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=32000,
+        hidden_size=4096,
+        intermediate_size=11008,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=None,
+        hidden_act="silu",
+        max_position_embeddings=2048,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=1,
+        eos_token_id=2,
+        pretraining_tp=1,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        attention_bias=False,
+        attention_dropout=0.0,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.pretraining_tp = pretraining_tp
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        self._rope_scaling_validation()
+        self.attention_bias = attention_bias
+        self.attention_dropout = attention_dropout
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+    def _rope_scaling_validation(self):
+        """
+        Validate the `rope_scaling` configuration.
+        """
+        if self.rope_scaling is None:
+            return
+
+        if not isinstance(self.rope_scaling, dict) or len(self.rope_scaling) != 2:
+            raise ValueError(
+                "`rope_scaling` must be a dictionary with with two fields, `type` and `factor`, "
+                f"got {self.rope_scaling}"
+            )
+        rope_scaling_type = self.rope_scaling.get("type", None)
+        rope_scaling_factor = self.rope_scaling.get("factor", None)
+        if rope_scaling_type is None or rope_scaling_type not in ["linear", "dynamic"]:
+            raise ValueError(
+                f"`rope_scaling`'s type field must be one of ['linear', 'dynamic'], got {rope_scaling_type}"
+            )
+        if rope_scaling_factor is None or not isinstance(rope_scaling_factor, float) or rope_scaling_factor <= 1.0:
+            raise ValueError(f"`rope_scaling`'s factor field must be a float > 1, got {rope_scaling_factor}")
+
+            
+class MplugOwlVisionConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`MplugOwlVisionModel`]. It is used to instantiate
+    a
+     mPLUG-Owl vision encoder according to the specified arguments, defining the model architecture. Instantiating a
+     configuration defaults will yield a similar configuration to that of the mPLUG-Owl
+     [x-plug/x_plug-llama-7b](https://huggingface.co/x-plug/x_plug-llama-7b) architecture.
+
+     Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+     documentation from [`PretrainedConfig`] for more information.
+
+     Args:
+         hidden_size (`int`, *optional*, defaults to 768):
+             Dimensionality of the encoder layers and the pooler layer.
+         intermediate_size (`int`, *optional*, defaults to 3072):
+             Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
+         num_hidden_layers (`int`, *optional*, defaults to 12):
+             Number of hidden layers in the Transformer encoder.
+         num_attention_heads (`int`, *optional*, defaults to 12):
+             Number of attention heads for each attention layer in the Transformer encoder.
+         image_size (`int`, *optional*, defaults to 224):
+             The size (resolution) of each image.
+         patch_size (`int`, *optional*, defaults to 32):
+             The size (resolution) of each patch.
+         hidden_act (`str` or `function`, *optional*, defaults to `"quick_gelu"`):
+             The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
+             `"relu"`, `"selu"` and `"gelu_new"` ``"quick_gelu"` are supported.
+         layer_norm_eps (`float`, *optional*, defaults to 1e-5):
+             The epsilon used by the layer normalization layers.
+         attention_dropout (`float`, *optional*, defaults to 0.0):
+             The dropout ratio for the attention probabilities.
+         initializer_range (`float`, *optional*, defaults to 0.02):
+             The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+         initializer_factor (`float`, *optional*, defaults to 1):
+             A factor for initializing all weight matrices (should be kept to 1, used internally for initialization
+             testing).
+
+
+     ```"""
+
+    model_type = "mplug_owl_vision_model"
+
+    def __init__(
+        self,
+        hidden_size=1024,
+        intermediate_size=4096,
+        projection_dim=768,
+        num_hidden_layers=24,
+        num_attention_heads=16,
+        num_channels=3,
+        image_size=1024,
+        patch_size=14,
+        hidden_act="quick_gelu",
+        layer_norm_eps=1e-6,
+        attention_dropout=0.0,
+        initializer_range=0.02,
+        initializer_factor=1.0,
+        use_flash_attn=False,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.projection_dim = projection_dim
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.num_channels = num_channels
+        self.patch_size = patch_size
+        self.image_size = image_size
+        self.initializer_range = initializer_range
+        self.initializer_factor = initializer_factor
+        self.attention_dropout = attention_dropout
+        self.layer_norm_eps = layer_norm_eps
+        self.hidden_act = hidden_act
+        self.use_flash_attn = use_flash_attn
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs) -> "PretrainedConfig":
+        config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)
+
+        # get the vision config dict if we are loading from MplugOwlConfig
+        if config_dict.get("model_type") == "mplug-owl":
+            config_dict = config_dict["vision_config"]
+
+        if "model_type" in config_dict and hasattr(cls, "model_type") and config_dict["model_type"] != cls.model_type:
+            logger.warning(
+                f"You are using a model of type {config_dict['model_type']} to instantiate a model of type "
+                f"{cls.model_type}. This is not supported for all configurations of models and can yield errors."
+            )
+
+        return cls.from_dict(config_dict, **kwargs)
+
+
+class MplugOwlVisualAbstractorConfig(PretrainedConfig):
+    model_type = "mplug_owl_visual_abstract"
+
+    def __init__(
+        self,
+        num_learnable_queries=64,
+        hidden_size=1024,
+        num_hidden_layers=6,
+        num_attention_heads=16,
+        intermediate_size=2816,
+        attention_probs_dropout_prob=0.,
+        initializer_range=0.02,
+        layer_norm_eps=1e-6,
+        encoder_hidden_size=1024,
+        grid_size=None,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.hidden_size = hidden_size
+        self.num_learnable_queries = num_learnable_queries
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.intermediate_size = intermediate_size
+        self.attention_probs_dropout_prob = attention_probs_dropout_prob
+        self.initializer_range = initializer_range
+        self.layer_norm_eps = layer_norm_eps
+        self.encoder_hidden_size = encoder_hidden_size
+        self.grid_size = grid_size if grid_size else 73
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs) -> "PretrainedConfig":
+        config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)
+
+        # get the visual_abstractor config dict if we are loading from MplugOwlConfig
+        if config_dict.get("model_type") == "mplug-owl":
+            config_dict = config_dict["abstractor_config"]
+
+        if "model_type" in config_dict and hasattr(cls, "model_type") and config_dict["model_type"] != cls.model_type:
+            logger.warning(
+                f"You are using a model of type {config_dict['model_type']} to instantiate a model of type "
+                f"{cls.model_type}. This is not supported for all configurations of models and can yield errors."
+            )
+
+        return cls.from_dict(config_dict, **kwargs)
+
+
+
+DEFAULT_VISUAL_CONFIG = {
+    "visual_model": MplugOwlVisionConfig().to_dict(),
+    "visual_abstractor": MplugOwlVisualAbstractorConfig().to_dict()
+}
+
+class MPLUGOwl2Config(LlamaConfig):
+    model_type = "mplug_owl2"
+    def __init__(self, visual_config=None, **kwargs):
+        if visual_config is None:
+            self.visual_config = DEFAULT_VISUAL_CONFIG
+        else:
+            self.visual_config = visual_config
+        
+        super().__init__(
+            **kwargs,
+        )
+        
+if __name__ == "__main__":
+    print(MplugOwlVisionConfig().to_dict())

modeling_attn_mask_utils.py

@@ -0,0 +1,275 @@
+# Copyright 2023 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from typing import List, Optional, Tuple, Union
+
+import torch
+
+
+class AttentionMaskConverter:
+    """
+    A utility attention mask class that allows one to:
+        - Create a causal 4d mask
+        - Create a causal 4d mask with slided window
+        - Convert a 2d attention mask (batch_size, query_length) to a 4d attention mask (batch_size, 1, query_length,
+          key_value_length) that can be multiplied with attention scores
+
+    Parameters:
+        is_causal (`bool`):
+            Whether the attention mask should be a uni-directional (causal) or bi-directional mask.
+
+        sliding_window (`int`, *optional*):
+            Optionally, the sliding window masks can be created if `sliding_window` is defined to a positive integer.
+    """
+
+    def __init__(self, is_causal: bool, sliding_window: Optional[int] = None):
+        self.is_causal = is_causal
+        self.sliding_window = sliding_window
+
+        if self.sliding_window is not None and self.sliding_window <= 0:
+            raise ValueError(
+                f"Make sure that when passing `sliding_window` that its value is a strictly positive integer, not `{self.sliding_window}`"
+            )
+
+    def to_causal_4d(
+        self,
+        batch_size: int,
+        query_length: int,
+        key_value_length: int,
+        dtype: torch.dtype = torch.float32,
+        device: Union[torch.device, "str"] = "cpu",
+    ) -> torch.Tensor:
+        """
+        Creates a causal 4D mask of (bsz, head_dim=1, query_length, key_value_length) shape and adds large negative
+        bias to upper right hand triangular matrix (causal mask).
+        """
+        if not self.is_causal:
+            raise ValueError(f"Please use `to_causal_4d` only if {self.__class__} has `is_causal` set to True.")
+
+        # If shape is not cached, create a new causal mask and cache it
+        input_shape = (batch_size, query_length)
+        past_key_values_length = key_value_length - query_length
+
+        # create causal mask
+        # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+        causal_4d_mask = None
+        if input_shape[-1] > 1 or self.sliding_window is not None:
+            causal_4d_mask = self._make_causal_mask(
+                input_shape,
+                dtype,
+                device=device,
+                past_key_values_length=past_key_values_length,
+                sliding_window=self.sliding_window,
+            )
+
+        return causal_4d_mask
+
+    def to_4d(
+        self,
+        attention_mask_2d: torch.Tensor,
+        query_length: int,
+        key_value_length: Optional[int] = None,
+        dtype: torch.dtype = torch.float32,
+    ) -> torch.Tensor:
+        """
+        Converts 2D attention mask to 4D attention mask by expanding mask to (bsz, head_dim=1, query_length,
+        key_value_length) shape and by adding a large negative bias to not-attended positions. If attention_mask is
+        causal, a causal mask will be added.
+        """
+        input_shape = (attention_mask_2d.shape[0], query_length)
+
+        # create causal mask
+        # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+        causal_4d_mask = None
+        if (input_shape[-1] > 1 or self.sliding_window is not None) and self.is_causal:
+            if key_value_length is None:
+                raise ValueError(
+                    "This attention mask converter is causal. Make sure to pass `key_value_length` to correctly create a causal mask."
+                )
+
+            past_key_values_length = key_value_length - query_length
+            causal_4d_mask = self._make_causal_mask(
+                input_shape,
+                dtype,
+                device=attention_mask_2d.device,
+                past_key_values_length=past_key_values_length,
+                sliding_window=self.sliding_window,
+            )
+        elif self.sliding_window is not None:
+            raise NotImplementedError("Sliding window is currently only implemented for causal masking")
+
+        # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+        expanded_attn_mask = self._expand_mask(attention_mask_2d, dtype, tgt_len=input_shape[-1]).to(
+            attention_mask_2d.device
+        )
+        expanded_4d_mask = expanded_attn_mask if causal_4d_mask is None else expanded_attn_mask + causal_4d_mask
+
+        return expanded_4d_mask
+
+    @staticmethod
+    def _make_causal_mask(
+        input_ids_shape: torch.Size,
+        dtype: torch.dtype,
+        device: torch.device,
+        past_key_values_length: int = 0,
+        sliding_window: Optional[int] = None,
+    ):
+        """
+        Make causal mask used for bi-directional self-attention.
+        """
+        bsz, tgt_len = input_ids_shape
+        mask = torch.full((tgt_len, tgt_len), torch.finfo(dtype).min, device=device)
+        mask_cond = torch.arange(mask.size(-1), device=device)
+        mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
+
+        mask = mask.to(dtype)
+
+        if past_key_values_length > 0:
+            mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
+
+        # add lower triangular sliding window mask if necessary
+        if sliding_window is not None:
+            diagonal = past_key_values_length - sliding_window + 1
+
+            context_mask = 1 - torch.triu(torch.ones_like(mask, dtype=torch.int), diagonal=diagonal)
+            mask.masked_fill_(context_mask.bool(), torch.finfo(dtype).min)
+
+        return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)
+
+    @staticmethod
+    def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
+        """
+        Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
+        """
+        bsz, src_len = mask.size()
+        tgt_len = tgt_len if tgt_len is not None else src_len
+
+        expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
+
+        inverted_mask = 1.0 - expanded_mask
+
+        return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)
+
+
+def _prepare_4d_causal_attention_mask(
+    attention_mask: Optional[torch.Tensor],
+    input_shape: Union[torch.Size, Tuple, List],
+    inputs_embeds: torch.Tensor,
+    past_key_values_length: int,
+    sliding_window: Optional[int] = None,
+):
+    """
+    Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape
+    `(batch_size, key_value_length)`
+
+    Args:
+        attention_mask (`torch.Tensor` or `None`):
+            A 2D attention mask of shape `(batch_size, key_value_length)`
+        input_shape (`tuple(int)` or `list(int)` or `torch.Size`):
+            The input shape should be a tuple that defines `(batch_size, query_length)`.
+        inputs_embeds (`torch.Tensor`):
+            The embedded inputs as a torch Tensor.
+        past_key_values_length (`int`):
+            The length of the key value cache.
+        sliding_window (`int`, *optional*):
+            If the model uses windowed attention, a sliding window should be passed.
+    """
+    attn_mask_converter = AttentionMaskConverter(is_causal=True, sliding_window=sliding_window)
+
+    key_value_length = input_shape[-1] + past_key_values_length
+
+    # 4d mask is passed through the layers
+    if attention_mask is not None:
+        attention_mask = attn_mask_converter.to_4d(
+            attention_mask, input_shape[-1], key_value_length, dtype=inputs_embeds.dtype
+        )
+    else:
+        attention_mask = attn_mask_converter.to_causal_4d(
+            input_shape[0], input_shape[-1], key_value_length, dtype=inputs_embeds.dtype, device=inputs_embeds.device
+        )
+
+    return attention_mask
+
+
+def _prepare_4d_attention_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
+    """
+    Creates a non-causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape
+    `(batch_size, key_value_length)`
+
+    Args:
+        mask (`torch.Tensor` or `None`):
+            A 2D attention mask of shape `(batch_size, key_value_length)`
+        dtype (`torch.dtype`):
+            The torch dtype the created mask shall have.
+        tgt_len (`int`):
+            The target length or query length the created mask shall have.
+    """
+    return AttentionMaskConverter._expand_mask(mask=mask, dtype=dtype, tgt_len=tgt_len)
+
+
+def _create_4d_causal_attention_mask(
+    input_shape: Union[torch.Size, Tuple, List],
+    dtype: torch.dtype,
+    device: torch.device,
+    past_key_values_length: int = 0,
+    sliding_window: Optional[int] = None,
+):
+    """
+    Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)`
+
+    Args:
+        input_shape (`tuple(int)` or `list(int)` or `torch.Size`):
+            The input shape should be a tuple that defines `(batch_size, query_length)`.
+        dtype (`torch.dtype`):
+            The torch dtype the created mask shall have.
+        device (`int`):
+            The torch device the created mask shall have.
+        sliding_window (`int`, *optional*):
+            If the model uses windowed attention, a sliding window should be passed.
+    """
+    attn_mask_converter = AttentionMaskConverter(is_causal=True, sliding_window=sliding_window)
+
+    key_value_length = past_key_values_length + input_shape[-1]
+    attention_mask = attn_mask_converter.to_causal_4d(
+        input_shape[0], input_shape[-1], key_value_length, dtype=dtype, device=device
+    )
+
+    return attention_mask
+
+def _prepare_4d_causal_attention_mask_for_sdpa(
+    attention_mask: Optional[torch.Tensor],
+    input_shape: Union[torch.Size, Tuple, List],
+    inputs_embeds: torch.Tensor,
+    past_key_values_length: int,
+    sliding_window: Optional[int] = None,
+):
+    """
+    Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape
+    `(batch_size, key_value_length)` for SDPA (Scaled Dot Product Attention).
+
+    Args:
+        attention_mask (`torch.Tensor` or `None`):
+            A 2D attention mask of shape `(batch_size, key_value_length)`
+        input_shape (`tuple(int)` or `list(int)` or `torch.Size`):
+            The input shape should be a tuple that defines `(batch_size, query_length)`.
+        inputs_embeds (`torch.Tensor`):
+            The embedded inputs as a torch Tensor.
+        past_key_values_length (`int`):
+            The length of the key value cache.
+        sliding_window (`int`, *optional*):
+            If the model uses windowed attention, a sliding window should be passed.
+    """
+    # For SDPA, we use the same implementation as the regular causal attention mask
+    return _prepare_4d_causal_attention_mask(
+        attention_mask, input_shape, inputs_embeds, past_key_values_length, sliding_window
+    )

modeling_llama2.py

@@ -0,0 +1,861 @@
+import math
+import warnings
+from functools import partial
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+from transformers.cache_utils import Cache
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.utils import logging
+from transformers.models.llama.modeling_llama import (
+    LlamaRotaryEmbedding,
+    LlamaLinearScalingRotaryEmbedding,
+    LlamaDynamicNTKScalingRotaryEmbedding,
+    apply_rotary_pos_emb,
+    repeat_kv,
+    LlamaMLP,
+    LlamaRMSNorm,
+    is_flash_attn_greater_or_equal_2_10,
+)
+
+logger = logging.get_logger(__name__)
+
+import copy
+import os
+import sys
+
+dir_path = os.path.dirname(os.path.realpath(__file__))
+sys.path.insert(0, dir_path)
+
+import transformers
+from transformers.models.llama.modeling_llama import *
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+from .modeling_attn_mask_utils import _prepare_4d_causal_attention_mask, _prepare_4d_causal_attention_mask_for_sdpa
+from .configuration_mplug_owl2 import LlamaConfig
+
+def _get_unpad_data(attention_mask):
+    seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
+    indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
+    max_seqlen_in_batch = seqlens_in_batch.max().item()
+    cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.torch.int32), (1, 0))
+    return (
+        indices,
+        cu_seqlens,
+        max_seqlen_in_batch,
+    )
+
+
+class MultiwayNetwork(nn.Module):
+
+    def __init__(self, module_provider, num_multiway=2):
+        super(MultiwayNetwork, self).__init__()
+
+        self.multiway = torch.nn.ModuleList([module_provider() for _ in range(num_multiway)])
+    
+    def forward(self, hidden_states, multiway_indices):
+
+        if len(self.multiway) == 1:
+            return self.multiway[0](hidden_states)
+
+        output_hidden_states = torch.empty_like(hidden_states)
+        
+        for idx, subway in enumerate(self.multiway):
+            local_indices = multiway_indices.eq(idx).nonzero(as_tuple=True)
+            hidden = hidden_states[local_indices].unsqueeze(1).contiguous()
+            if hidden.numel():
+                output = subway(hidden)
+                if isinstance(output, tuple):
+                    output = output[0]
+                output = output.squeeze(1)
+                output_hidden_states[local_indices] = output
+        
+        return output_hidden_states.contiguous()
+    
+
+class LlamaAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: LlamaConfig, layer_idx: Optional[int] = None):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        if layer_idx is None:
+            logger.warning_once(
+                f"Instantiating {self.__class__.__name__} without passing `layer_idx` is not recommended and will "
+                "to errors during the forward call, if caching is used. Please make sure to provide a `layer_idx` "
+                "when creating this class."
+            )
+            
+        self.attention_dropout = config.attention_dropout
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.hidden_size // self.num_heads
+        self.num_key_value_heads = config.num_key_value_heads
+        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
+        self.max_position_embeddings = config.max_position_embeddings
+        self.rope_theta = config.rope_theta
+        self.is_causal = True
+
+        if (self.head_dim * self.num_heads) != self.hidden_size:
+            raise ValueError(
+                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
+                f" and `num_heads`: {self.num_heads})."
+            )
+        self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=config.attention_bias)
+        self.k_proj = MultiwayNetwork(module_provider=partial(
+            nn.Linear, in_features=self.hidden_size, out_features=self.num_key_value_heads * self.head_dim, bias=config.attention_bias)
+        )
+        self.v_proj = MultiwayNetwork(module_provider=partial(
+            nn.Linear, in_features=self.hidden_size, out_features=self.num_key_value_heads * self.head_dim, bias=config.attention_bias)
+        )
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=config.attention_bias)
+        self._init_rope()
+
+    def _init_rope(self):
+        if self.config.rope_scaling is None:
+            self.rotary_emb = LlamaRotaryEmbedding(
+                self.head_dim,
+                max_position_embeddings=self.max_position_embeddings,
+                base=self.rope_theta,
+            )
+        else:
+            scaling_type = self.config.rope_scaling["type"]
+            scaling_factor = self.config.rope_scaling["factor"]
+            if scaling_type == "linear":
+                self.rotary_emb = LlamaLinearScalingRotaryEmbedding(
+                    self.head_dim,
+                    max_position_embeddings=self.max_position_embeddings,
+                    scaling_factor=scaling_factor,
+                    base=self.rope_theta,
+                )
+            elif scaling_type == "dynamic":
+                self.rotary_emb = LlamaDynamicNTKScalingRotaryEmbedding(
+                    self.head_dim,
+                    max_position_embeddings=self.max_position_embeddings,
+                    scaling_factor=scaling_factor,
+                    base=self.rope_theta,
+                )
+            else:
+                raise ValueError(f"Unknown RoPE scaling type {scaling_type}")
+
+    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        modality_indicators: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        padding_mask: Optional[torch.LongTensor] = None,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states, )
+        key_states = self.k_proj(hidden_states, modality_indicators)
+        value_states = self.v_proj(hidden_states, modality_indicators)
+
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value[0].shape[-2]
+        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+
+        if past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = torch.cat([past_key_value[0], key_states], dim=2)
+            value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+        past_key_value = (key_states, value_states) if use_cache else None
+
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+            attn_weights = attn_weights + attention_mask
+
+        # upcast attention to fp32
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.transpose(1, 2).contiguous()
+
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+    
+
+class LlamaFlashAttention2(LlamaAttention):
+    """
+    Llama flash attention module. This module inherits from `LlamaAttention` as the weights of the module stays
+    untouched. The only required change would be on the forward pass where it needs to correctly call the public API of
+    flash attention and deal with padding tokens in case the input contains any of them.
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
+        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
+        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
+        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        modality_indicators: torch.Tensor,
+        attention_mask: Optional[torch.LongTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        **kwargs,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        # LlamaFlashAttention2 attention does not support output_attentions
+        if "padding_mask" in kwargs:
+            warnings.warn(
+                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
+            )
+
+            # overwrite attention_mask with padding_mask
+            attention_mask = kwargs.pop("padding_mask")
+
+        output_attentions = False
+
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states, modality_indicators)
+        value_states = self.v_proj(hidden_states, modality_indicators)
+
+        # Flash attention requires the input to have the shape
+        # batch_size x seq_length x head_dim x hidden_dim
+        # therefore we just need to keep the original shape
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx)
+        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+
+        if past_key_value is not None:
+            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        # TODO: These transpose are quite inefficient but Flash Attention requires the layout [batch_size, sequence_length, num_heads, head_dim]. We would need to refactor the KV cache
+        # to be able to avoid many of these transpose/reshape/view.
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        dropout_rate = self.attention_dropout if self.training else 0.0
+
+        # In PEFT, usually we cast the layer norms in float32 for training stability reasons
+        # therefore the input hidden states gets silently casted in float32. Hence, we need
+        # cast them back in the correct dtype just to be sure everything works as expected.
+        # This might slowdown training & inference so it is recommended to not cast the LayerNorms
+        # in fp32. (LlamaRMSNorm handles it correctly)
+
+        input_dtype = query_states.dtype
+        if input_dtype == torch.float32:
+            if torch.is_autocast_enabled():
+                target_dtype = torch.get_autocast_gpu_dtype()
+            # Handle the case where the model is quantized
+            elif hasattr(self.config, "_pre_quantization_dtype"):
+                target_dtype = self.config._pre_quantization_dtype
+            else:
+                target_dtype = self.q_proj.weight.dtype
+
+            logger.warning_once(
+                f"The input hidden states seems to be silently casted in float32, this might be related to"
+                f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
+                f" {target_dtype}."
+            )
+
+            query_states = query_states.to(target_dtype)
+            key_states = key_states.to(target_dtype)
+            value_states = value_states.to(target_dtype)
+
+        attn_output = self._flash_attention_forward(
+            query_states, key_states, value_states, attention_mask, q_len, dropout=dropout_rate
+        )
+
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size).contiguous()
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+    def _flash_attention_forward(
+        self, query_states, key_states, value_states, attention_mask, query_length, dropout=0.0, softmax_scale=None
+    ):
+        """
+        Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
+        first unpad the input, then computes the attention scores and pad the final attention scores.
+
+        Args:
+            query_states (`torch.Tensor`):
+                Input query states to be passed to Flash Attention API
+            key_states (`torch.Tensor`):
+                Input key states to be passed to Flash Attention API
+            value_states (`torch.Tensor`):
+                Input value states to be passed to Flash Attention API
+            attention_mask (`torch.Tensor`):
+                The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
+                position of padding tokens and 1 for the position of non-padding tokens.
+            dropout (`int`, *optional*):
+                Attention dropout
+            softmax_scale (`float`, *optional*):
+                The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
+        """
+        if not self._flash_attn_uses_top_left_mask:
+            causal = self.is_causal
+        else:
+            # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
+            causal = self.is_causal and query_length != 1
+
+        # Contains at least one padding token in the sequence
+        if attention_mask is not None:
+            batch_size = query_states.shape[0]
+            query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
+                query_states, key_states, value_states, attention_mask, query_length
+            )
+
+            cu_seqlens_q, cu_seqlens_k = cu_seq_lens
+            max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
+
+            attn_output_unpad = flash_attn_varlen_func(
+                query_states,
+                key_states,
+                value_states,
+                cu_seqlens_q=cu_seqlens_q,
+                cu_seqlens_k=cu_seqlens_k,
+                max_seqlen_q=max_seqlen_in_batch_q,
+                max_seqlen_k=max_seqlen_in_batch_k,
+                dropout_p=dropout,
+                softmax_scale=softmax_scale,
+                causal=causal,
+            )
+
+            attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
+        else:
+            attn_output = flash_attn_func(
+                query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal
+            )
+
+        return attn_output
+
+    def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
+        indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
+        batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
+
+        key_layer = index_first_axis(
+            key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
+        )
+        value_layer = index_first_axis(
+            value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
+        )
+        if query_length == kv_seq_len:
+            query_layer = index_first_axis(
+                query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim), indices_k
+            )
+            cu_seqlens_q = cu_seqlens_k
+            max_seqlen_in_batch_q = max_seqlen_in_batch_k
+            indices_q = indices_k
+        elif query_length == 1:
+            max_seqlen_in_batch_q = 1
+            cu_seqlens_q = torch.arange(
+                batch_size + 1, dtype=torch.int32, device=query_layer.device
+            )  # There is a memcpy here, that is very bad.
+            indices_q = cu_seqlens_q[:-1]
+            query_layer = query_layer.squeeze(1)
+        else:
+            # The -q_len: slice assumes left padding.
+            attention_mask = attention_mask[:, -query_length:]
+            query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
+
+        return (
+            query_layer,
+            key_layer,
+            value_layer,
+            indices_q,
+            (cu_seqlens_q, cu_seqlens_k),
+            (max_seqlen_in_batch_q, max_seqlen_in_batch_k),
+        )
+
+
+class LlamaSdpaAttention(LlamaAttention):
+    """
+    Llama attention module using torch.nn.functional.scaled_dot_product_attention. This module inherits from
+    `LlamaAttention` as the weights of the module stays untouched. The only changes are on the forward pass to adapt to
+    SDPA API.
+    """
+
+    # Adapted from LlamaAttention.forward
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        modality_indicators: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        if output_attentions:
+            # TODO: Improve this warning with e.g. `model.config.attn_implementation = "manual"` once this is implemented.
+            logger.warning_once(
+                "LlamaModel is using LlamaSdpaAttention, but `torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to the manual attention implementation, "
+                'but specifying the manual implementation will be required from Transformers version v5.0.0 onwards. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
+            )
+            return super().forward(
+                hidden_states=hidden_states,
+                modality_indicators=modality_indicators,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_value,
+                output_attentions=output_attentions,
+                use_cache=use_cache,
+            )
+
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states, modality_indicators)
+        value_states = self.v_proj(hidden_states, modality_indicators)
+
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx)
+        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+
+        if past_key_value is not None:
+            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+
+        # SDPA with memory-efficient backend is currently (torch==2.1.2) bugged with non-contiguous inputs with custom attn_mask,
+        # Reference: https://github.com/pytorch/pytorch/issues/112577.
+        if query_states.device.type == "cuda" and attention_mask is not None:
+            query_states = query_states.contiguous()
+            key_states = key_states.contiguous()
+            value_states = value_states.contiguous()
+
+        attn_output = torch.nn.functional.scaled_dot_product_attention(
+            query_states,
+            key_states,
+            value_states,
+            attn_mask=attention_mask,
+            dropout_p=self.attention_dropout if self.training else 0.0,
+            # The q_len > 1 is necessary to match with AttentionMaskConverter.to_causal_4d that does not create a causal mask in case q_len == 1.
+            is_causal=self.is_causal and attention_mask is None and q_len > 1,
+        )
+
+        attn_output = attn_output.transpose(1, 2).contiguous()
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+        attn_output = self.o_proj(attn_output)
+
+        return attn_output, None, past_key_value
+
+
+
+LLAMA_ATTENTION_CLASSES = {
+    "eager": LlamaAttention,
+    "flash_attention_2": LlamaFlashAttention2,
+    "sdpa": LlamaSdpaAttention,
+}
+
+class LlamaDecoderLayer(nn.Module):
+    def __init__(self, config: LlamaConfig, layer_idx):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.self_attn = LlamaAttention(config=config)
+        # Get attention implementation with fallback to 'eager'
+        attn_implementation = getattr(config, '_attn_implementation', 'eager')
+        if attn_implementation in LLAMA_ATTENTION_CLASSES:
+            self.self_attn = LLAMA_ATTENTION_CLASSES[attn_implementation](config=config, layer_idx=layer_idx)
+        else:
+            # Fallback to eager implementation
+            self.self_attn = LLAMA_ATTENTION_CLASSES['eager'](config=config, layer_idx=layer_idx)
+        self.mlp = LlamaMLP(config)
+        self.input_layernorm = MultiwayNetwork(module_provider=partial(
+            LlamaRMSNorm, hidden_size=config.hidden_size, eps=config.rms_norm_eps
+        ))
+        self.post_attention_layernorm = MultiwayNetwork(module_provider=partial(
+            LlamaRMSNorm, hidden_size=config.hidden_size, eps=config.rms_norm_eps
+        ))
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        modality_indicators: torch.Tensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+        """
+
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states, modality_indicators)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            modality_indicators=modality_indicators,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states, modality_indicators)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs
+
+
+def model_forward(
+    self,
+    input_ids: torch.LongTensor = None,
+    modality_indicators: torch.Tensor = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[List[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+) -> Union[Tuple, BaseModelOutputWithPast]:
+    output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+    output_hidden_states = (
+        output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+    )
+    use_cache = use_cache if use_cache is not None else self.config.use_cache
+
+    return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+    # Set attention implementation attributes if not already set
+    if not hasattr(self, '_use_flash_attention_2'):
+        self._use_flash_attention_2 = getattr(self.config, '_attn_implementation', 'eager') == 'flash_attention_2'
+    if not hasattr(self, '_use_sdpa'):
+        self._use_sdpa = getattr(self.config, '_attn_implementation', 'eager') == 'sdpa'
+
+    # retrieve input_ids and inputs_embeds
+    if input_ids is not None and inputs_embeds is not None:
+        raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
+    elif input_ids is not None:
+        batch_size, seq_length = input_ids.shape
+    elif inputs_embeds is not None:
+        batch_size, seq_length, _ = inputs_embeds.shape
+    else:
+        raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")
+    
+    seq_length_with_past = seq_length
+    past_key_values_length = 0
+
+    if past_key_values is not None:
+        past_key_values_length = past_key_values[0][0].shape[2]
+        seq_length_with_past = seq_length_with_past + past_key_values_length
+
+    if position_ids is None:
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+        position_ids = torch.arange(
+            past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
+        )
+        position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+    else:
+        position_ids = position_ids.view(-1, seq_length).long()
+
+    if inputs_embeds is None:
+        inputs_embeds = self.embed_tokens(input_ids)
+    # embed positions
+    if attention_mask is None:
+        attention_mask = torch.ones(
+            (batch_size, seq_length_with_past), dtype=torch.bool, device=inputs_embeds.device
+        )
+        
+    if self._use_flash_attention_2:
+            # 2d mask is passed through the layers
+            attention_mask = attention_mask if (attention_mask is not None and 0 in attention_mask) else None
+    elif self._use_sdpa and not output_attentions:
+            # output_attentions=True can not be supported when using SDPA, and we fall back on
+            # the manual implementation that requires a 4D causal mask in all cases.
+            attention_mask = _prepare_4d_causal_attention_mask_for_sdpa(
+                attention_mask,
+                (batch_size, seq_length),
+                inputs_embeds,
+                past_key_values_length,
+            )
+    else:
+            # 4d mask is passed through the layers
+            attention_mask = _prepare_4d_causal_attention_mask(
+                attention_mask, (batch_size, seq_length), inputs_embeds, past_key_values_length
+            )
+
+    hidden_states = inputs_embeds
+
+    if self.gradient_checkpointing and self.training:
+        if use_cache:
+            logger.warning_once(
+                "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+            )
+            use_cache = False
+
+    # decoder layers
+    all_hidden_states = () if output_hidden_states else None
+    all_self_attns = () if output_attentions else None
+    next_decoder_cache = () if use_cache else None
+
+    for idx, decoder_layer in enumerate(self.layers):
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        past_key_value = past_key_values[idx] if past_key_values is not None else None
+
+        if self.gradient_checkpointing and self.training:
+
+            def create_custom_forward(module):
+                def custom_forward(*inputs):
+                    # None for past_key_value
+                    return module(*inputs, past_key_value, output_attentions)
+
+                return custom_forward
+
+            layer_outputs = torch.utils.checkpoint.checkpoint(
+                create_custom_forward(decoder_layer),
+                hidden_states,
+                modality_indicators,
+                attention_mask,
+                position_ids,
+            )
+        else:
+            layer_outputs = decoder_layer(
+                hidden_states,
+                modality_indicators=modality_indicators,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_value,
+                output_attentions=output_attentions,
+                use_cache=use_cache,
+            )
+
+        hidden_states = layer_outputs[0]
+
+        if use_cache:
+            next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
+
+        if output_attentions:
+            all_self_attns += (layer_outputs[1],)
+
+    hidden_states = self.norm(hidden_states)
+
+    # add hidden states from the last decoder layer
+    if output_hidden_states:
+        all_hidden_states += (hidden_states,)
+
+    next_cache = next_decoder_cache if use_cache else None
+    if not return_dict:
+        return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
+    return BaseModelOutputWithPast(
+        last_hidden_state=hidden_states,
+        past_key_values=next_cache,
+        hidden_states=all_hidden_states,
+        attentions=all_self_attns,
+    )
+
+
+def causal_model_forward(
+    self,
+    input_ids: torch.LongTensor = None,
+    modality_indicators: torch.Tensor = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[List[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    labels: Optional[torch.LongTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+) -> Union[Tuple, CausalLMOutputWithPast]:
+    r"""
+    Args:
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+    Returns:
+
+    Example:
+
+    ```python
+    >>> from transformers import AutoTokenizer, LlamaForCausalLM
+
+    >>> model = LlamaForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS)
+    >>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)
+
+    >>> prompt = "Hey, are you conscious? Can you talk to me?"
+    >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+    >>> # Generate
+    >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+    >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+    ```"""
+
+    output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+    output_hidden_states = (
+        output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+    )
+    return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+    # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+    outputs = self.model(
+        input_ids=input_ids,
+        modality_indicators=modality_indicators,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        past_key_values=past_key_values,
+        inputs_embeds=inputs_embeds,
+        use_cache=use_cache,
+        output_attentions=output_attentions,
+        output_hidden_states=output_hidden_states,
+        return_dict=return_dict,
+    )
+
+    hidden_states = outputs[0]
+    if self.config.pretraining_tp > 1:
+        lm_head_slices = self.lm_head.weight.split(self.vocab_size // self.config.pretraining_tp, dim=0)
+        logits = [F.linear(hidden_states, lm_head_slices[i]) for i in range(self.config.pretraining_tp)]
+        logits = torch.cat(logits, dim=-1)
+    else:
+        logits = self.lm_head(hidden_states)
+    logits = logits.float()
+
+    loss = None
+    if labels is not None:
+        # Shift so that tokens < n predict n
+        shift_logits = logits[..., :-1, :].contiguous()
+        shift_labels = labels[..., 1:].contiguous()
+        # Flatten the tokens
+        loss_fct = CrossEntropyLoss()
+        shift_logits = shift_logits.view(-1, self.config.vocab_size)
+        shift_labels = shift_labels.view(-1)
+        # Enable model parallelism
+        shift_labels = shift_labels.to(shift_logits.device)
+        loss = loss_fct(shift_logits, shift_labels)
+
+    if not return_dict:
+        output = (logits,) + outputs[1:]
+        return (loss,) + output if loss is not None else output
+
+    return CausalLMOutputWithPast(
+        loss=loss,
+        logits=logits,
+        past_key_values=outputs.past_key_values,
+        hidden_states=outputs.hidden_states,
+        attentions=outputs.attentions,
+    )
+
+def replace_llama_modality_adaptive():
+    transformers.models.llama.configuration_llama.LlamaConfig = LlamaConfig
+    transformers.models.llama.modeling_llama.LlamaAttention = LlamaAttention
+    transformers.models.llama.modeling_llama.LlamaFlashAttention2 = LlamaFlashAttention2
+    transformers.models.llama.modeling_llama.LlamaSdpaAttention = LlamaSdpaAttention
+    transformers.models.llama.modeling_llama.LlamaDecoderLayer = LlamaDecoderLayer
+    transformers.models.llama.modeling_llama.LlamaModel.forward = model_forward
+    transformers.models.llama.modeling_llama.LlamaForCausalLM.forward = causal_model_forward
+
+    
+if __name__ == "__main__":
+    replace_llama_modality_adaptive()
+    config = transformers.LlamaConfig.from_pretrained('/cpfs01/shared/public/test/vicuna-7b-v1.5/')
+    model = transformers.LlamaForCausalLM(config)
+    print(model)

modeling_mplug_owl2_huggingface.py

@@ -0,0 +1,571 @@
+#    Copyright 2023 Haotian Liu & Qinghao Ye (Modified from LLaVA)
+#
+#    Licensed under the Apache License, Version 2.0 (the "License");
+#    you may not use this file except in compliance with the License.
+#    You may obtain a copy of the License at
+#
+#        http://www.apache.org/licenses/LICENSE-2.0
+#
+#    Unless required by applicable law or agreed to in writing, software
+#    distributed under the License is distributed on an "AS IS" BASIS,
+#    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#    See the License for the specific language governing permissions and
+#    limitations under the License.
+
+import os
+import sys
+from abc import ABC, abstractmethod
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from torch.nn import CrossEntropyLoss
+from PIL import Image
+
+dir_path = os.path.dirname(os.path.realpath(__file__))
+sys.path.insert(0, dir_path)
+
+from transformers import (
+    AutoConfig,
+    AutoModelForCausalLM,
+    AutoTokenizer,
+    LlamaForCausalLM,
+    LlamaModel,
+)
+from transformers.modeling_outputs import CausalLMOutputWithPast
+from transformers.models.clip.image_processing_clip import CLIPImageProcessor
+
+from .configuration_mplug_owl2 import (
+    MPLUGOwl2Config,
+    MplugOwlVisionConfig,
+    MplugOwlVisualAbstractorConfig
+)
+from .modeling_attn_mask_utils import _prepare_4d_causal_attention_mask  # Force copy
+from .modeling_llama2 import replace_llama_modality_adaptive
+from .visual_encoder import MplugOwlVisionModel, MplugOwlVisualAbstractorModel
+
+IGNORE_INDEX = -100
+IMAGE_TOKEN_INDEX = -200
+DEFAULT_IMAGE_TOKEN = "<|image|>"
+
+
+def tokenizer_image_token(prompt, tokenizer, image_token_index=IMAGE_TOKEN_INDEX, return_tensors=None):
+    prompt_chunks = [tokenizer(chunk).input_ids for chunk in prompt.split(DEFAULT_IMAGE_TOKEN)]
+
+    def insert_separator(X, sep):
+        return [ele for sublist in zip(X, [sep]*len(X)) for ele in sublist][:-1]
+
+    input_ids = []
+    offset = 0
+    if len(prompt_chunks) > 0 and len(prompt_chunks[0]) > 0 and prompt_chunks[0][0] == tokenizer.bos_token_id:
+        offset = 1
+        input_ids.append(prompt_chunks[0][0])
+
+    for x in insert_separator(prompt_chunks, [image_token_index] * (offset + 1)):
+        input_ids.extend(x[offset:])
+
+    if return_tensors is not None:
+        if return_tensors == 'pt':
+            return torch.tensor(input_ids, dtype=torch.long)
+        raise ValueError(f'Unsupported tensor type: {return_tensors}')
+    return input_ids
+
+
+def expand2square(pil_img, background_color):
+    width, height = pil_img.size
+    if width == height:
+        return pil_img
+    elif width > height:
+        result = Image.new(pil_img.mode, (width, width), background_color)
+        result.paste(pil_img, (0, (width - height) // 2))
+        return result
+    else:
+        result = Image.new(pil_img.mode, (height, height), background_color)
+        result.paste(pil_img, ((height - width) // 2, 0))
+        return result
+
+
+class MPLUGOwl2MetaModel:
+    def __init__(self, config):
+        super(MPLUGOwl2MetaModel, self).__init__(config)
+        self.vision_model = MplugOwlVisionModel(
+            MplugOwlVisionConfig(**config.visual_config["visual_model"])
+        )
+        self.visual_abstractor = MplugOwlVisualAbstractorModel(
+            MplugOwlVisualAbstractorConfig(**config.visual_config["visual_abstractor"]),
+            config.hidden_size,
+        )
+
+    def get_vision_tower(self):
+        vision_model = getattr(self, "vision_model", None)
+        if type(vision_model) is list:
+            vision_model = vision_model[0]
+        return vision_model
+
+    def get_visual_abstractor(self):
+        visual_abstractor = getattr(self, "visual_abstractor", None)
+        if type(visual_abstractor) is list:
+            visual_abstractor = visual_abstractor[0]
+        return visual_abstractor
+
+
+class MPLUGOwl2MetaForCausalLM(ABC):
+    @abstractmethod
+    def get_model(self):
+        pass
+
+    def encode_images(self, images):
+        image_features = self.get_model().vision_model(images).last_hidden_state
+        image_features = (
+            self.get_model()
+            .visual_abstractor(encoder_hidden_states=image_features)
+            .last_hidden_state
+        )
+        return image_features
+
+    def prepare_inputs_labels_for_multimodal(
+        self, input_ids, attention_mask, past_key_values, labels, images
+    ):
+        if images is None or input_ids.shape[1] == 1:
+            if (
+                past_key_values is not None
+                and images is not None
+                and input_ids.shape[1] == 1
+            ):
+                attention_mask = torch.ones(
+                    (attention_mask.shape[0], past_key_values[-1][-1].shape[-2] + 1),
+                    dtype=attention_mask.dtype,
+                    device=attention_mask.device,
+                )
+            multiway_indices = torch.zeros_like(input_ids).long().to(self.device)
+            return (
+                input_ids,
+                multiway_indices,
+                attention_mask,
+                past_key_values,
+                None,
+                labels,
+            )
+
+        if type(images) is list or images.ndim == 5:
+            concat_images = torch.cat([image for image in images], dim=0)
+            image_features = self.encode_images(concat_images)
+            split_sizes = [image.shape[0] for image in images]
+            image_features = torch.split(image_features, split_sizes, dim=0)
+            image_features = [x.flatten(0, 1) for x in image_features]
+        else:
+            image_features = self.encode_images(images)
+
+        new_input_embeds = []
+        new_modality_indicators = []
+        new_labels = [] if labels is not None else None
+        cur_image_idx = 0
+        for batch_idx, cur_input_ids in enumerate(input_ids):
+            if (cur_input_ids == IMAGE_TOKEN_INDEX).sum() == 0:
+                half_len = cur_input_ids.shape[0] // 2
+                cur_image_features = image_features[cur_image_idx]
+                cur_input_embeds_1 = self.get_model().embed_tokens(
+                    cur_input_ids[:half_len]
+                )
+                cur_input_embeds_2 = self.get_model().embed_tokens(
+                    cur_input_ids[half_len:]
+                )
+                cur_input_embeds = torch.cat(
+                    [cur_input_embeds_1, cur_image_features[0:0], cur_input_embeds_2],
+                    dim=0,
+                )
+                new_input_embeds.append(cur_input_embeds)
+
+                cur_modality_indicators = (
+                    torch.zeros(len(cur_input_embeds)).long().to(self.device)
+                )
+                new_modality_indicators.append(cur_modality_indicators)
+                if labels is not None:
+                    new_labels.append(labels[batch_idx])
+                cur_image_idx += 1
+                continue
+            image_token_indices = torch.where(cur_input_ids == IMAGE_TOKEN_INDEX)[0]
+            cur_new_input_embeds = []
+            cur_modality_indicators = []
+            if labels is not None:
+                cur_labels = labels[batch_idx]
+                cur_new_labels = []
+                assert cur_labels.shape == cur_input_ids.shape
+            while image_token_indices.numel() > 0:
+                cur_image_features = image_features[cur_image_idx]
+                image_token_start = image_token_indices[0]
+                cur_new_input_embeds.append(
+                    self.get_model().embed_tokens(cur_input_ids[:image_token_start])
+                )
+                cur_new_input_embeds.append(cur_image_features)
+
+                cur_modality_indicators.append(
+                    torch.zeros(len(cur_input_ids[:image_token_start])).long()
+                )
+                cur_modality_indicators.append(
+                    torch.ones(len(cur_image_features)).long()
+                )
+
+                if labels is not None:
+                    cur_new_labels.append(cur_labels[:image_token_start])
+                    cur_new_labels.append(
+                        torch.full(
+                            (cur_image_features.shape[0],),
+                            IGNORE_INDEX,
+                            device=labels.device,
+                            dtype=labels.dtype,
+                        )
+                    )
+                    cur_labels = cur_labels[image_token_start + 1 :]
+                cur_image_idx += 1
+                cur_input_ids = cur_input_ids[image_token_start + 1 :]
+                image_token_indices = torch.where(cur_input_ids == IMAGE_TOKEN_INDEX)[0]
+            if cur_input_ids.numel() > 0:
+                cur_new_input_embeds.append(
+                    self.get_model().embed_tokens(cur_input_ids)
+                )
+                cur_modality_indicators.append(torch.zeros(len(cur_input_ids)).long())
+                if labels is not None:
+                    cur_new_labels.append(cur_labels)
+            cur_new_input_embeds = [
+                x.to(device=self.device) for x in cur_new_input_embeds
+            ]
+            cur_new_input_embeds = torch.cat(cur_new_input_embeds, dim=0)
+            new_input_embeds.append(cur_new_input_embeds)
+
+            cur_modality_indicators = [
+                x.to(device=self.device) for x in cur_modality_indicators
+            ]
+            cur_modality_indicators = torch.cat(cur_modality_indicators, dim=0)
+            new_modality_indicators.append(cur_modality_indicators)
+
+            if labels is not None:
+                cur_new_labels = torch.cat(cur_new_labels, dim=0)
+                new_labels.append(cur_new_labels)
+
+        if any(x.shape != new_input_embeds[0].shape for x in new_input_embeds):
+            max_len = max(x.shape[0] for x in new_input_embeds)
+
+            new_input_embeds_align = []
+            for cur_new_embed in new_input_embeds:
+                cur_new_embed = torch.cat(
+                    (
+                        cur_new_embed,
+                        torch.zeros(
+                            (max_len - cur_new_embed.shape[0], cur_new_embed.shape[1]),
+                            dtype=cur_new_embed.dtype,
+                            device=cur_new_embed.device,
+                        ),
+                    ),
+                    dim=0,
+                )
+                new_input_embeds_align.append(cur_new_embed)
+            new_input_embeds = torch.stack(new_input_embeds_align, dim=0)
+
+            new_modality_indicators_align = []
+            for cur_modality_indicator in new_modality_indicators:
+                cur_new_embed = torch.cat(
+                    (
+                        cur_modality_indicator,
+                        torch.zeros(
+                            max_len - cur_modality_indicator.shape[0],
+                            dtype=cur_modality_indicator.dtype,
+                            device=cur_modality_indicator.device,
+                        ),
+                    ),
+                    dim=0,
+                )
+                new_modality_indicators_align.append(cur_new_embed)
+            new_modality_indicators = torch.stack(new_modality_indicators_align, dim=0)
+
+            if labels is not None:
+                new_labels_align = []
+                _new_labels = new_labels
+                for cur_new_label in new_labels:
+                    cur_new_label = torch.cat(
+                        (
+                            cur_new_label,
+                            torch.full(
+                                (max_len - cur_new_label.shape[0],),
+                                IGNORE_INDEX,
+                                dtype=cur_new_label.dtype,
+                                device=cur_new_label.device,
+                            ),
+                        ),
+                        dim=0,
+                    )
+                    new_labels_align.append(cur_new_label)
+                new_labels = torch.stack(new_labels_align, dim=0)
+
+            if attention_mask is not None:
+                new_attention_mask = []
+                for cur_attention_mask, cur_new_labels, cur_new_labels_align in zip(
+                    attention_mask, _new_labels, new_labels
+                ):
+                    new_attn_mask_pad_left = torch.full(
+                        (cur_new_labels.shape[0] - labels.shape[1],),
+                        True,
+                        dtype=attention_mask.dtype,
+                        device=attention_mask.device,
+                    )
+                    new_attn_mask_pad_right = torch.full(
+                        (cur_new_labels_align.shape[0] - cur_new_labels.shape[0],),
+                        False,
+                        dtype=attention_mask.dtype,
+                        device=attention_mask.device,
+                    )
+                    cur_new_attention_mask = torch.cat(
+                        (
+                            new_attn_mask_pad_left,
+                            cur_attention_mask,
+                            new_attn_mask_pad_right,
+                        ),
+                        dim=0,
+                    )
+                    new_attention_mask.append(cur_new_attention_mask)
+                attention_mask = torch.stack(new_attention_mask, dim=0)
+                assert attention_mask.shape == new_labels.shape
+        else:
+            new_input_embeds = torch.stack(new_input_embeds, dim=0)
+            new_modality_indicators = torch.stack(new_modality_indicators, dim=0)
+            if labels is not None:
+                new_labels = torch.stack(new_labels, dim=0)
+
+            if attention_mask is not None:
+                new_attn_mask_pad_left = torch.full(
+                    (
+                        attention_mask.shape[0],
+                        new_input_embeds.shape[1] - input_ids.shape[1],
+                    ),
+                    True,
+                    dtype=attention_mask.dtype,
+                    device=attention_mask.device,
+                )
+                attention_mask = torch.cat(
+                    (new_attn_mask_pad_left, attention_mask), dim=1
+                )
+                assert attention_mask.shape == new_input_embeds.shape[:2]
+        return (
+            None,
+            new_modality_indicators,
+            attention_mask,
+            past_key_values,
+            new_input_embeds,
+            new_labels,
+        )
+
+
+class MPLUGOwl2LlamaModel(MPLUGOwl2MetaModel, LlamaModel):
+    config_class = MPLUGOwl2Config
+
+    def __init__(self, config: MPLUGOwl2Config):
+        super(MPLUGOwl2LlamaModel, self).__init__(config)
+
+
+class MPLUGOwl2LlamaForCausalLM(LlamaForCausalLM, MPLUGOwl2MetaForCausalLM):
+    config_class = MPLUGOwl2Config
+
+    def __init__(self, config):
+        super(LlamaForCausalLM, self).__init__(config)
+        self.model = MPLUGOwl2LlamaModel(config)
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Tokenizer and image processor will be initialized lazily in score()
+        self._tokenizer = None
+        self._image_processor = None
+        self._preferential_ids = None
+
+        self.post_init()
+
+    def _init_processors(self):
+        """Lazily initialize tokenizer and image processor from the model's directory."""
+        if self._tokenizer is None:
+            # Use the model's name_or_path from config, fallback to HF repo name
+            model_path = getattr(self.config, '_name_or_path', None)
+            if model_path is None or model_path == './' or not model_path.startswith(('/', 'http', 'mapo80')):
+                model_path = "mapo80/DeQA-Doc-Sharpness"
+            self._tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
+            self._image_processor = CLIPImageProcessor.from_pretrained(model_path)
+            self._preferential_ids = [id_[1] for id_ in self._tokenizer(
+                ["excellent", "good", "fair", "poor", "bad"]
+            )["input_ids"]]
+
+    @property
+    def tokenizer(self):
+        self._init_processors()
+        return self._tokenizer
+
+    @property
+    def image_processor(self):
+        self._init_processors()
+        return self._image_processor
+
+    @property
+    def preferential_ids_(self):
+        self._init_processors()
+        return self._preferential_ids
+
+    def get_model(self):
+        return self.model
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        images: Optional[torch.FloatTensor] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+        (
+            input_ids,
+            modality_indicators,
+            attention_mask,
+            past_key_values,
+            inputs_embeds,
+            labels,
+        ) = self.prepare_inputs_labels_for_multimodal(
+            input_ids, attention_mask, past_key_values, labels, images
+        )
+
+        outputs = self.model(
+            input_ids=input_ids,
+            modality_indicators=modality_indicators,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        hidden_states = outputs[0]
+        logits = self.lm_head(hidden_states)
+
+        loss = None
+        if labels is not None:
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def score(
+        self,
+        images: List[Image.Image],
+        task_: str = "quality",
+        input_: str = "image",
+    ) -> torch.Tensor:
+        """
+        Score images based on quality assessment.
+
+        Args:
+            images: List of PIL Images to score
+            task_: Type of assessment (default: "quality")
+            input_: Input type - "image" or "video" (default: "image")
+
+        Returns:
+            torch.Tensor: Quality scores (1-5 scale)
+        """
+        if not hasattr(self, "weight_tensor"):
+            self.weight_tensor = torch.Tensor([5., 4., 3., 2., 1.]).half().to(self.device)
+
+        prompt = "USER: How would you rate the {} of this {}?\n<|image|>\nASSISTANT: The {} of the {} is".format(
+            task_, input_, task_, input_
+        )
+
+        if input_ == "image":
+            # Process single images
+            images = [expand2square(img, tuple(int(x*255) for x in self.image_processor.image_mean)) for img in images]
+            input_ids = tokenizer_image_token(prompt, self.tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt').unsqueeze(0).to(self.device)
+
+            with torch.inference_mode():
+                image_tensor = self.image_processor.preprocess(images, return_tensors="pt")["pixel_values"].half().to(self.device)
+                output_logits = self(
+                    input_ids=input_ids.repeat(image_tensor.shape[0], 1),
+                    images=image_tensor
+                )["logits"][:, -1, self.preferential_ids_]
+
+                return torch.softmax(output_logits, -1) @ self.weight_tensor
+        else:
+            # Process videos (list of frame sequences)
+            video = [[expand2square(frame, tuple(int(x*255) for x in self.image_processor.image_mean)) for frame in vid] for vid in images]
+            input_ids = tokenizer_image_token(prompt, self.tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt').unsqueeze(0).to(self.device)
+
+            with torch.inference_mode():
+                video_tensors = [
+                    self.image_processor.preprocess(vid, return_tensors="pt")["pixel_values"].half().to(self.device)
+                    for vid in video
+                ]
+                output_logits = self(
+                    input_ids=input_ids.repeat(len(video_tensors), 1),
+                    images=video_tensors
+                )["logits"][:, -1, self.preferential_ids_]
+
+                return torch.softmax(output_logits, -1) @ self.weight_tensor
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        past_key_values=None,
+        attention_mask=None,
+        inputs_embeds=None,
+        images=None,
+        **kwargs,
+    ):
+        if past_key_values:
+            input_ids = input_ids[:, -1:]
+
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "attention_mask": attention_mask,
+                "images": images,
+            }
+        )
+        return model_inputs
+
+
+AutoConfig.register("mplug_owl2", MPLUGOwl2Config)
+AutoModelForCausalLM.register(MPLUGOwl2Config, MPLUGOwl2LlamaForCausalLM)
+
+replace_llama_modality_adaptive()

preprocessor_config.json

@@ -0,0 +1,18 @@
+{
+  "crop_size": {
+    "height": 448,
+    "width": 448
+  },
+  "do_center_crop": true,
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "image_mean": [0.48145466, 0.4578275, 0.40821073],
+  "image_processor_type": "CLIPImageProcessor",
+  "image_std": [0.26862954, 0.26130258, 0.27577711],
+  "resample": 3,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "shortest_edge": 448
+  }
+}

pytorch_model-00001-of-00004.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:721e2348e05ef81b5cc9534a94b9caba3699586a37cd643aad6dd6c14a9b7041
+size 4981108780

pytorch_model-00002-of-00004.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94954006aba2f33919e4082bc65fc5b5091f73998e9915f85cb71e832d6ee96d
+size 4920293259

pytorch_model-00003-of-00004.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:639b15209838cf92a603a4f8667cdec11582c01e4fa7aba87ed916dce074ab76
+size 4989532454

pytorch_model-00004-of-00004.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c358f2e9c72bf180194543d9c167d23bd22ca3aa0c9d689afa21d505ca8ab8c
+size 1518469929

pytorch_model.bin.index.json

@@ -0,0 +1,869 @@
+{
+  "metadata": {
+    "total_size": 16409100288
+  },
+  "weight_map": {
+    "lm_head.weight": "pytorch_model-00004-of-00004.bin",
+    "model.embed_tokens.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.input_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.input_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.mlp.down_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.mlp.gate_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.mlp.up_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.post_attention_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.post_attention_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.self_attn.k_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.self_attn.k_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.self_attn.o_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.self_attn.q_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.self_attn.v_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.0.self_attn.v_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.input_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.input_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.mlp.down_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.mlp.gate_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.mlp.up_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.post_attention_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.post_attention_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.self_attn.k_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.self_attn.k_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.self_attn.o_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.self_attn.q_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.self_attn.v_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.1.self_attn.v_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.10.input_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.10.input_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.10.mlp.down_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.10.mlp.gate_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.10.mlp.up_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.10.post_attention_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.10.post_attention_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.10.self_attn.k_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.10.self_attn.k_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.10.self_attn.o_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.10.self_attn.q_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.10.self_attn.v_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.10.self_attn.v_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.input_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.input_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.mlp.down_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.mlp.gate_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.mlp.up_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.post_attention_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.post_attention_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.self_attn.k_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.self_attn.k_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.self_attn.o_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.self_attn.q_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.self_attn.v_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.11.self_attn.v_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.input_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.input_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.mlp.down_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.mlp.gate_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.mlp.up_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.post_attention_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.post_attention_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.self_attn.k_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.self_attn.k_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.self_attn.o_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.self_attn.q_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.self_attn.v_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.12.self_attn.v_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.input_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.input_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.mlp.down_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.mlp.gate_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.mlp.up_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.post_attention_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.post_attention_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.self_attn.k_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.self_attn.k_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.self_attn.o_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.self_attn.q_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.self_attn.v_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.13.self_attn.v_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.input_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.input_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.mlp.down_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.mlp.gate_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.mlp.up_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.post_attention_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.post_attention_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.self_attn.k_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.self_attn.k_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.self_attn.o_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.self_attn.q_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.self_attn.v_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.14.self_attn.v_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.input_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.input_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.mlp.down_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.mlp.gate_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.mlp.up_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.post_attention_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.post_attention_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.self_attn.k_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.self_attn.k_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.self_attn.o_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.self_attn.q_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.self_attn.v_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.15.self_attn.v_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.input_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.input_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.mlp.down_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.mlp.gate_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.mlp.up_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.post_attention_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.post_attention_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.self_attn.k_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.self_attn.k_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.self_attn.o_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.self_attn.q_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.self_attn.v_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.16.self_attn.v_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.input_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.input_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.mlp.down_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.mlp.gate_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.mlp.up_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.post_attention_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.post_attention_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.self_attn.k_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.self_attn.k_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.self_attn.o_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.self_attn.q_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.self_attn.v_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.17.self_attn.v_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.input_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.input_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.mlp.down_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.mlp.gate_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.mlp.up_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.post_attention_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.post_attention_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.self_attn.k_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.self_attn.k_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.self_attn.o_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.self_attn.q_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.self_attn.v_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.18.self_attn.v_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.input_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.input_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.mlp.down_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.mlp.gate_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.mlp.up_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.post_attention_layernorm.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.post_attention_layernorm.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.self_attn.k_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.self_attn.k_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.self_attn.o_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.self_attn.q_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.self_attn.v_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.19.self_attn.v_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.2.input_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.2.input_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.2.mlp.down_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.2.mlp.gate_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.2.mlp.up_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.2.post_attention_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.2.post_attention_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.2.self_attn.k_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.2.self_attn.k_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.2.self_attn.o_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.2.self_attn.q_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.2.self_attn.v_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.2.self_attn.v_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.20.input_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.20.input_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.20.mlp.down_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.20.mlp.gate_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.20.mlp.up_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.20.post_attention_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.20.post_attention_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.20.self_attn.k_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.20.self_attn.k_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.20.self_attn.o_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.20.self_attn.q_proj.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.20.self_attn.v_proj.multiway.0.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.20.self_attn.v_proj.multiway.1.weight": "pytorch_model-00002-of-00004.bin",
+    "model.layers.21.input_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.21.input_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.21.mlp.down_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.21.mlp.gate_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.21.mlp.up_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.21.post_attention_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.21.post_attention_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.21.self_attn.k_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.21.self_attn.k_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.21.self_attn.o_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.21.self_attn.q_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.21.self_attn.v_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.21.self_attn.v_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.input_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.input_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.mlp.down_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.mlp.gate_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.mlp.up_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.post_attention_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.post_attention_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.self_attn.k_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.self_attn.k_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.self_attn.o_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.self_attn.q_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.self_attn.v_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.22.self_attn.v_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.input_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.input_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.mlp.down_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.mlp.gate_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.mlp.up_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.post_attention_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.post_attention_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.self_attn.k_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.self_attn.k_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.self_attn.o_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.self_attn.q_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.self_attn.v_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.23.self_attn.v_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.input_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.input_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.mlp.down_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.mlp.gate_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.mlp.up_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.post_attention_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.post_attention_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.self_attn.k_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.self_attn.k_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.self_attn.o_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.self_attn.q_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.self_attn.v_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.24.self_attn.v_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.input_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.input_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.mlp.down_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.mlp.gate_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.mlp.up_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.post_attention_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.post_attention_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.self_attn.k_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.self_attn.k_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.self_attn.o_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.self_attn.q_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.self_attn.v_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.25.self_attn.v_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.input_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.input_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.mlp.down_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.mlp.gate_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.mlp.up_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.post_attention_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.post_attention_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.self_attn.k_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.self_attn.k_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.self_attn.o_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.self_attn.q_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.self_attn.v_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.26.self_attn.v_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.input_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.input_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.mlp.down_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.mlp.gate_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.mlp.up_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.post_attention_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.post_attention_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.self_attn.k_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.self_attn.k_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.self_attn.o_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.self_attn.q_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.self_attn.v_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.27.self_attn.v_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.input_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.input_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.mlp.down_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.mlp.gate_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.mlp.up_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.post_attention_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.post_attention_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.self_attn.k_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.self_attn.k_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.self_attn.o_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.self_attn.q_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.self_attn.v_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.28.self_attn.v_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.input_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.input_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.mlp.down_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.mlp.gate_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.mlp.up_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.post_attention_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.post_attention_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.self_attn.k_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.self_attn.k_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.self_attn.o_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.self_attn.q_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.self_attn.v_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.29.self_attn.v_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.3.input_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.3.input_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.3.mlp.down_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.3.mlp.gate_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.3.mlp.up_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.3.post_attention_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.3.post_attention_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.3.self_attn.k_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.3.self_attn.k_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.3.self_attn.o_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.3.self_attn.q_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.3.self_attn.v_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.3.self_attn.v_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.30.input_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.30.input_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.30.mlp.down_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.30.mlp.gate_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.30.mlp.up_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.30.post_attention_layernorm.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.30.post_attention_layernorm.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.30.self_attn.k_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.30.self_attn.k_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.30.self_attn.o_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.30.self_attn.q_proj.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.30.self_attn.v_proj.multiway.0.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.30.self_attn.v_proj.multiway.1.weight": "pytorch_model-00003-of-00004.bin",
+    "model.layers.31.input_layernorm.multiway.0.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.31.input_layernorm.multiway.1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.31.mlp.down_proj.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.31.mlp.gate_proj.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.31.mlp.up_proj.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.31.post_attention_layernorm.multiway.0.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.31.post_attention_layernorm.multiway.1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.31.self_attn.k_proj.multiway.0.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.31.self_attn.k_proj.multiway.1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.31.self_attn.o_proj.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.31.self_attn.q_proj.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.31.self_attn.v_proj.multiway.0.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.31.self_attn.v_proj.multiway.1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.layers.4.input_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.4.input_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.4.mlp.down_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.4.mlp.gate_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.4.mlp.up_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.4.post_attention_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.4.post_attention_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.4.self_attn.k_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.4.self_attn.k_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.4.self_attn.o_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.4.self_attn.q_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.4.self_attn.v_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.4.self_attn.v_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.input_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.input_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.mlp.down_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.mlp.gate_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.mlp.up_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.post_attention_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.post_attention_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.self_attn.k_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.self_attn.k_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.self_attn.o_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.self_attn.q_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.self_attn.v_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.5.self_attn.v_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.input_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.input_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.mlp.down_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.mlp.gate_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.mlp.up_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.post_attention_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.post_attention_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.self_attn.k_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.self_attn.k_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.self_attn.o_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.self_attn.q_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.self_attn.v_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.6.self_attn.v_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.input_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.input_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.mlp.down_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.mlp.gate_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.mlp.up_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.post_attention_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.post_attention_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.self_attn.k_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.self_attn.k_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.self_attn.o_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.self_attn.q_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.self_attn.v_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.7.self_attn.v_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.input_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.input_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.mlp.down_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.mlp.gate_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.mlp.up_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.post_attention_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.post_attention_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.self_attn.k_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.self_attn.k_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.self_attn.o_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.self_attn.q_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.self_attn.v_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.8.self_attn.v_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.input_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.input_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.mlp.down_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.mlp.gate_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.mlp.up_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.post_attention_layernorm.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.post_attention_layernorm.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.self_attn.k_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.self_attn.k_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.self_attn.o_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.self_attn.q_proj.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.self_attn.v_proj.multiway.0.weight": "pytorch_model-00001-of-00004.bin",
+    "model.layers.9.self_attn.v_proj.multiway.1.weight": "pytorch_model-00001-of-00004.bin",
+    "model.norm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.embeddings.cls_token": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.embeddings.patch_embed.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.embeddings.position_embedding": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.embeddings.pre_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.embeddings.pre_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.0.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.0.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.0.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.0.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.0.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.0.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.0.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.0.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.0.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.0.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.0.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.0.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.1.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.1.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.1.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.1.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.1.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.1.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.1.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.1.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.1.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.1.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.1.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.1.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.10.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.10.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.10.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.10.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.10.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.10.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.10.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.10.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.10.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.10.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.10.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.10.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.11.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.11.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.11.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.11.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.11.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.11.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.11.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.11.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.11.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.11.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.11.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.11.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.12.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.12.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.12.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.12.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.12.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.12.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.12.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.12.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.12.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.12.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.12.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.12.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.13.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.13.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.13.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.13.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.13.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.13.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.13.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.13.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.13.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.13.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.13.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.13.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.14.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.14.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.14.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.14.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.14.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.14.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.14.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.14.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.14.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.14.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.14.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.14.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.15.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.15.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.15.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.15.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.15.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.15.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.15.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.15.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.15.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.15.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.15.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.15.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.16.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.16.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.16.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.16.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.16.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.16.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.16.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.16.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.16.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.16.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.16.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.16.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.17.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.17.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.17.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.17.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.17.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.17.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.17.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.17.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.17.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.17.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.17.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.17.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.18.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.18.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.18.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.18.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.18.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.18.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.18.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.18.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.18.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.18.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.18.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.18.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.19.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.19.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.19.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.19.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.19.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.19.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.19.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.19.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.19.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.19.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.19.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.19.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.2.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.2.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.2.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.2.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.2.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.2.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.2.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.2.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.2.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.2.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.2.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.2.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.20.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.20.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.20.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.20.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.20.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.20.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.20.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.20.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.20.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.20.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.20.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.20.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.21.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.21.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.21.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.21.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.21.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.21.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.21.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.21.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.21.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.21.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.21.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.21.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.22.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.22.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.22.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.22.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.22.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.22.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.22.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.22.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.22.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.22.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.22.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.22.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.23.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.23.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.23.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.23.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.23.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.23.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.23.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.23.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.23.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.23.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.23.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.23.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.3.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.3.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.3.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.3.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.3.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.3.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.3.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.3.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.3.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.3.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.3.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.3.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.4.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.4.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.4.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.4.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.4.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.4.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.4.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.4.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.4.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.4.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.4.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.4.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.5.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.5.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.5.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.5.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.5.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.5.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.5.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.5.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.5.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.5.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.5.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.5.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.6.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.6.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.6.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.6.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.6.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.6.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.6.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.6.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.6.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.6.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.6.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.6.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.7.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.7.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.7.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.7.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.7.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.7.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.7.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.7.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.7.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.7.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.7.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.7.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.8.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.8.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.8.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.8.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.8.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.8.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.8.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.8.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.8.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.8.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.8.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.8.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.9.input_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.9.input_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.9.mlp.fc1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.9.mlp.fc1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.9.mlp.fc2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.9.mlp.fc2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.9.post_attention_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.9.post_attention_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.9.self_attn.dense.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.9.self_attn.dense.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.9.self_attn.query_key_value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.encoder.layers.9.self_attn.query_key_value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.post_layernorm.bias": "pytorch_model-00004-of-00004.bin",
+    "model.vision_model.post_layernorm.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.attention.k_pos_embed": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.attention.key.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.attention.key.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.attention.q_pos_embed": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.attention.query.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.attention.query.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.attention.value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.attention.value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.norm1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.norm1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.normk.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.normk.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.output.mlp.ffn_ln.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.output.mlp.ffn_ln.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.output.mlp.w1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.output.mlp.w1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.output.mlp.w2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.output.mlp.w2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.output.mlp.w3.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.output.mlp.w3.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.output.norm2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.output.norm2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.output.out_proj.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.0.crossattention.output.out_proj.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.attention.k_pos_embed": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.attention.key.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.attention.key.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.attention.q_pos_embed": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.attention.query.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.attention.query.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.attention.value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.attention.value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.norm1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.norm1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.normk.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.normk.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.output.mlp.ffn_ln.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.output.mlp.ffn_ln.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.output.mlp.w1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.output.mlp.w1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.output.mlp.w2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.output.mlp.w2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.output.mlp.w3.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.output.mlp.w3.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.output.norm2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.output.norm2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.output.out_proj.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.1.crossattention.output.out_proj.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.attention.k_pos_embed": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.attention.key.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.attention.key.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.attention.q_pos_embed": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.attention.query.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.attention.query.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.attention.value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.attention.value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.norm1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.norm1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.normk.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.normk.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.output.mlp.ffn_ln.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.output.mlp.ffn_ln.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.output.mlp.w1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.output.mlp.w1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.output.mlp.w2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.output.mlp.w2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.output.mlp.w3.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.output.mlp.w3.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.output.norm2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.output.norm2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.output.out_proj.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.2.crossattention.output.out_proj.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.attention.k_pos_embed": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.attention.key.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.attention.key.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.attention.q_pos_embed": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.attention.query.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.attention.query.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.attention.value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.attention.value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.norm1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.norm1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.normk.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.normk.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.output.mlp.ffn_ln.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.output.mlp.ffn_ln.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.output.mlp.w1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.output.mlp.w1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.output.mlp.w2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.output.mlp.w2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.output.mlp.w3.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.output.mlp.w3.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.output.norm2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.output.norm2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.output.out_proj.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.3.crossattention.output.out_proj.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.attention.k_pos_embed": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.attention.key.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.attention.key.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.attention.q_pos_embed": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.attention.query.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.attention.query.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.attention.value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.attention.value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.norm1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.norm1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.normk.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.normk.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.output.mlp.ffn_ln.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.output.mlp.ffn_ln.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.output.mlp.w1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.output.mlp.w1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.output.mlp.w2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.output.mlp.w2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.output.mlp.w3.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.output.mlp.w3.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.output.norm2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.output.norm2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.output.out_proj.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.4.crossattention.output.out_proj.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.attention.k_pos_embed": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.attention.key.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.attention.key.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.attention.q_pos_embed": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.attention.query.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.attention.query.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.attention.value.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.attention.value.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.norm1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.norm1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.normk.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.normk.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.output.mlp.ffn_ln.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.output.mlp.ffn_ln.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.output.mlp.w1.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.output.mlp.w1.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.output.mlp.w2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.output.mlp.w2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.output.mlp.w3.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.output.mlp.w3.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.output.norm2.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.output.norm2.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.output.out_proj.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.encoder.layers.5.crossattention.output.out_proj.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.query_embeds": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.visual_fc.bias": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.visual_fc.weight": "pytorch_model-00004-of-00004.bin",
+    "model.visual_abstractor.vit_eos": "pytorch_model-00004-of-00004.bin"
+  }
+}

runs/Jun19_10-38-29_antfcutrn-kmaker-033145119198/events.out.tfevents.1750300731.antfcutrn-kmaker-033145119198.18341.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:139a2814d9a266e78a1f3e035a1ac2b48cd2966a690ac7b1a7427fd63911a891
+size 62257

special_tokens_map.json

@@ -0,0 +1,24 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<unk>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723

tokenizer_config.json

@@ -0,0 +1,42 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": false,
+  "model_max_length": 2048,
+  "pad_token": "<unk>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

trainer_state.json

@@ -0,0 +1,2010 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 500,
+  "global_step": 330,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 0.1273,
+      "step": 1
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 0.1438,
+      "step": 2
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6e-06,
+      "loss": 0.1274,
+      "step": 3
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 0.0856,
+      "step": 4
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 1e-05,
+      "loss": 0.0508,
+      "step": 5
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 1.2e-05,
+      "loss": 0.066,
+      "step": 6
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 1.4e-05,
+      "loss": 0.0804,
+      "step": 7
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 1.6000000000000003e-05,
+      "loss": 0.0526,
+      "step": 8
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 1.8e-05,
+      "loss": 0.0692,
+      "step": 9
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 2e-05,
+      "loss": 0.0485,
+      "step": 10
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 1.999951808959328e-05,
+      "loss": 0.0363,
+      "step": 11
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 1.9998072404820648e-05,
+      "loss": 0.0303,
+      "step": 12
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1.9995663085020215e-05,
+      "loss": 0.0527,
+      "step": 13
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 1.9992290362407232e-05,
+      "loss": 0.0485,
+      "step": 14
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 1.9987954562051724e-05,
+      "loss": 0.0726,
+      "step": 15
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 1.998265610184716e-05,
+      "loss": 0.0379,
+      "step": 16
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 1.997639549247016e-05,
+      "loss": 0.0427,
+      "step": 17
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1.9969173337331283e-05,
+      "loss": 0.0405,
+      "step": 18
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 1.9960990332516875e-05,
+      "loss": 0.0359,
+      "step": 19
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 1.995184726672197e-05,
+      "loss": 0.0284,
+      "step": 20
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1.9941745021174284e-05,
+      "loss": 0.033,
+      "step": 21
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 1.9930684569549265e-05,
+      "loss": 0.0328,
+      "step": 22
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 1.991866697787626e-05,
+      "loss": 0.022,
+      "step": 23
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 1.990569340443577e-05,
+      "loss": 0.0374,
+      "step": 24
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1.989176509964781e-05,
+      "loss": 0.0214,
+      "step": 25
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 1.9876883405951378e-05,
+      "loss": 0.0329,
+      "step": 26
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 1.9861049757675087e-05,
+      "loss": 0.0114,
+      "step": 27
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 1.9844265680898917e-05,
+      "loss": 0.0174,
+      "step": 28
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1.982653279330712e-05,
+      "loss": 0.0339,
+      "step": 29
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 1.9807852804032306e-05,
+      "loss": 0.0477,
+      "step": 30
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 1.9788227513490724e-05,
+      "loss": 0.0377,
+      "step": 31
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 1.9767658813208725e-05,
+      "loss": 0.0261,
+      "step": 32
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 1.974614868564045e-05,
+      "loss": 0.0397,
+      "step": 33
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 1.9723699203976768e-05,
+      "loss": 0.0316,
+      "step": 34
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 1.9700312531945444e-05,
+      "loss": 0.0323,
+      "step": 35
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1.96759909236026e-05,
+      "loss": 0.0264,
+      "step": 36
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 1.9650736723115476e-05,
+      "loss": 0.0204,
+      "step": 37
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 1.9624552364536472e-05,
+      "loss": 0.02,
+      "step": 38
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 1.9597440371568576e-05,
+      "loss": 0.0189,
+      "step": 39
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 1.956940335732209e-05,
+      "loss": 0.0277,
+      "step": 40
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 1.9540444024062807e-05,
+      "loss": 0.0321,
+      "step": 41
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1.9510565162951538e-05,
+      "loss": 0.04,
+      "step": 42
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 1.9479769653775107e-05,
+      "loss": 0.0308,
+      "step": 43
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 1.944806046466878e-05,
+      "loss": 0.0235,
+      "step": 44
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 1.941544065183021e-05,
+      "loss": 0.0185,
+      "step": 45
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1.9381913359224844e-05,
+      "loss": 0.0477,
+      "step": 46
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 1.9347481818282927e-05,
+      "loss": 0.0353,
+      "step": 47
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 1.9312149347588035e-05,
+      "loss": 0.0347,
+      "step": 48
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 1.9275919352557242e-05,
+      "loss": 0.023,
+      "step": 49
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 1.9238795325112867e-05,
+      "loss": 0.0752,
+      "step": 50
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 1.920078084334595e-05,
+      "loss": 0.0305,
+      "step": 51
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1.916187957117136e-05,
+      "loss": 0.0216,
+      "step": 52
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 1.9122095257974676e-05,
+      "loss": 0.0129,
+      "step": 53
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 1.9081431738250815e-05,
+      "loss": 0.0225,
+      "step": 54
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9039892931234434e-05,
+      "loss": 0.022,
+      "step": 55
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.8997482840522218e-05,
+      "loss": 0.0419,
+      "step": 56
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.895420555368697e-05,
+      "loss": 0.0308,
+      "step": 57
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.891006524188368e-05,
+      "loss": 0.0246,
+      "step": 58
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8865066159447468e-05,
+      "loss": 0.0446,
+      "step": 59
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.881921264348355e-05,
+      "loss": 0.0255,
+      "step": 60
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8772509113449243e-05,
+      "loss": 0.0247,
+      "step": 61
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.8724960070727974e-05,
+      "loss": 0.0189,
+      "step": 62
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.8676570098195443e-05,
+      "loss": 0.0241,
+      "step": 63
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.862734385977792e-05,
+      "loss": 0.0285,
+      "step": 64
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.8577286100002723e-05,
+      "loss": 0.0266,
+      "step": 65
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.8526401643540924e-05,
+      "loss": 0.0156,
+      "step": 66
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.8474695394742345e-05,
+      "loss": 0.0188,
+      "step": 67
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.8422172337162865e-05,
+      "loss": 0.0172,
+      "step": 68
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.8368837533084092e-05,
+      "loss": 0.019,
+      "step": 69
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.8314696123025456e-05,
+      "loss": 0.0165,
+      "step": 70
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.825975332524873e-05,
+      "loss": 0.0179,
+      "step": 71
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.8204014435255136e-05,
+      "loss": 0.0199,
+      "step": 72
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.8147484825274895e-05,
+      "loss": 0.0377,
+      "step": 73
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.8090169943749477e-05,
+      "loss": 0.0196,
+      "step": 74
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.803207531480645e-05,
+      "loss": 0.0103,
+      "step": 75
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.797320653772707e-05,
+      "loss": 0.0237,
+      "step": 76
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.7913569286406606e-05,
+      "loss": 0.0238,
+      "step": 77
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.785316930880745e-05,
+      "loss": 0.0182,
+      "step": 78
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.779201242640517e-05,
+      "loss": 0.0171,
+      "step": 79
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.773010453362737e-05,
+      "loss": 0.0256,
+      "step": 80
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.7667451597285617e-05,
+      "loss": 0.0196,
+      "step": 81
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.7604059656000313e-05,
+      "loss": 0.0221,
+      "step": 82
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.7539934819618696e-05,
+      "loss": 0.0172,
+      "step": 83
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 1.747508326862597e-05,
+      "loss": 0.0192,
+      "step": 84
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 1.7409511253549592e-05,
+      "loss": 0.0141,
+      "step": 85
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1.7343225094356857e-05,
+      "loss": 0.0226,
+      "step": 86
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 1.727623117984575e-05,
+      "loss": 0.0158,
+      "step": 87
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 1.720853596702919e-05,
+      "loss": 0.0158,
+      "step": 88
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 1.7140145980512684e-05,
+      "loss": 0.0186,
+      "step": 89
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1.7071067811865477e-05,
+      "loss": 0.0099,
+      "step": 90
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 1.7001308118985237e-05,
+      "loss": 0.0174,
+      "step": 91
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 1.6930873625456362e-05,
+      "loss": 0.0254,
+      "step": 92
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 1.685977111990193e-05,
+      "loss": 0.0173,
+      "step": 93
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 1.678800745532942e-05,
+      "loss": 0.0132,
+      "step": 94
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 1.6715589548470187e-05,
+      "loss": 0.0118,
+      "step": 95
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 1.664252437911282e-05,
+      "loss": 0.0159,
+      "step": 96
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.6568818989430416e-05,
+      "loss": 0.0165,
+      "step": 97
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.6494480483301836e-05,
+      "loss": 0.0167,
+      "step": 98
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.641951602562703e-05,
+      "loss": 0.0083,
+      "step": 99
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 1.6343932841636455e-05,
+      "loss": 0.0229,
+      "step": 100
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 1.6267738216194698e-05,
+      "loss": 0.0202,
+      "step": 101
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 1.6190939493098344e-05,
+      "loss": 0.0273,
+      "step": 102
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 1.6113544074368166e-05,
+      "loss": 0.0117,
+      "step": 103
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 1.6035559419535714e-05,
+      "loss": 0.0133,
+      "step": 104
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 1.5956993044924334e-05,
+      "loss": 0.018,
+      "step": 105
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.5877852522924733e-05,
+      "loss": 0.0175,
+      "step": 106
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.579814548126514e-05,
+      "loss": 0.0255,
+      "step": 107
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.5717879602276123e-05,
+      "loss": 0.0223,
+      "step": 108
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 1.5637062622150168e-05,
+      "loss": 0.0218,
+      "step": 109
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.5555702330196024e-05,
+      "loss": 0.0102,
+      "step": 110
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 1.547380656808797e-05,
+      "loss": 0.0127,
+      "step": 111
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 1.5391383229110005e-05,
+      "loss": 0.017,
+      "step": 112
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 1.5308440257395095e-05,
+      "loss": 0.0184,
+      "step": 113
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 1.5224985647159489e-05,
+      "loss": 0.0077,
+      "step": 114
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 1.5141027441932217e-05,
+      "loss": 0.0139,
+      "step": 115
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 1.5056573733779848e-05,
+      "loss": 0.0109,
+      "step": 116
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 1.4971632662526545e-05,
+      "loss": 0.0146,
+      "step": 117
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 1.4886212414969551e-05,
+      "loss": 0.0187,
+      "step": 118
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 1.4800321224090114e-05,
+      "loss": 0.019,
+      "step": 119
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 1.4713967368259981e-05,
+      "loss": 0.0103,
+      "step": 120
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 1.4627159170443504e-05,
+      "loss": 0.0164,
+      "step": 121
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 1.4539904997395468e-05,
+      "loss": 0.0131,
+      "step": 122
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 1.4452213258854684e-05,
+      "loss": 0.0172,
+      "step": 123
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 1.436409240673342e-05,
+      "loss": 0.0125,
+      "step": 124
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 1.4275550934302822e-05,
+      "loss": 0.0102,
+      "step": 125
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 1.4186597375374283e-05,
+      "loss": 0.0134,
+      "step": 126
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 1.4097240303476955e-05,
+      "loss": 0.0067,
+      "step": 127
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 1.4007488331031409e-05,
+      "loss": 0.0096,
+      "step": 128
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 1.391735010851956e-05,
+      "loss": 0.0154,
+      "step": 129
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 1.3826834323650899e-05,
+      "loss": 0.0188,
+      "step": 130
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 1.3735949700525164e-05,
+      "loss": 0.008,
+      "step": 131
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 1.3644704998791501e-05,
+      "loss": 0.0139,
+      "step": 132
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 1.3553109012804162e-05,
+      "loss": 0.0111,
+      "step": 133
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 1.346117057077493e-05,
+      "loss": 0.0136,
+      "step": 134
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 1.3368898533922202e-05,
+      "loss": 0.012,
+      "step": 135
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 1.3276301795616937e-05,
+      "loss": 0.0181,
+      "step": 136
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 1.3183389280525497e-05,
+      "loss": 0.012,
+      "step": 137
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 1.3090169943749475e-05,
+      "loss": 0.0085,
+      "step": 138
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 1.2996652769962567e-05,
+      "loss": 0.0107,
+      "step": 139
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 1.2902846772544625e-05,
+      "loss": 0.015,
+      "step": 140
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 1.2808760992712923e-05,
+      "loss": 0.0152,
+      "step": 141
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 1.2714404498650743e-05,
+      "loss": 0.0126,
+      "step": 142
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 1.2619786384633374e-05,
+      "loss": 0.0086,
+      "step": 143
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 1.252491577015158e-05,
+      "loss": 0.0121,
+      "step": 144
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 1.242980179903264e-05,
+      "loss": 0.0146,
+      "step": 145
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 1.2334453638559057e-05,
+      "loss": 0.0146,
+      "step": 146
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 1.2238880478584987e-05,
+      "loss": 0.014,
+      "step": 147
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 1.2143091530650508e-05,
+      "loss": 0.0207,
+      "step": 148
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 1.2047096027093798e-05,
+      "loss": 0.0166,
+      "step": 149
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 1.1950903220161286e-05,
+      "loss": 0.015,
+      "step": 150
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 1.185452238111591e-05,
+      "loss": 0.0115,
+      "step": 151
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 1.1757962799343548e-05,
+      "loss": 0.0081,
+      "step": 152
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 1.1661233781457655e-05,
+      "loss": 0.0164,
+      "step": 153
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 1.156434465040231e-05,
+      "loss": 0.0108,
+      "step": 154
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 1.1467304744553618e-05,
+      "loss": 0.021,
+      "step": 155
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 1.1370123416819683e-05,
+      "loss": 0.0187,
+      "step": 156
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 1.1272810033739134e-05,
+      "loss": 0.0178,
+      "step": 157
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 1.1175373974578378e-05,
+      "loss": 0.0195,
+      "step": 158
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 1.1077824630427593e-05,
+      "loss": 0.016,
+      "step": 159
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 1.098017140329561e-05,
+      "loss": 0.0226,
+      "step": 160
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 1.0882423705203698e-05,
+      "loss": 0.0187,
+      "step": 161
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 1.0784590957278452e-05,
+      "loss": 0.0122,
+      "step": 162
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 1.0686682588843737e-05,
+      "loss": 0.0073,
+      "step": 163
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 1.058870803651189e-05,
+      "loss": 0.0102,
+      "step": 164
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 1.0490676743274181e-05,
+      "loss": 0.0076,
+      "step": 165
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 1.0392598157590687e-05,
+      "loss": 0.011,
+      "step": 166
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 1.0294481732479635e-05,
+      "loss": 0.02,
+      "step": 167
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 1.0196336924606282e-05,
+      "loss": 0.0186,
+      "step": 168
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 1.0098173193371498e-05,
+      "loss": 0.0133,
+      "step": 169
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 1e-05,
+      "loss": 0.0128,
+      "step": 170
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 9.901826806628505e-06,
+      "loss": 0.0067,
+      "step": 171
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 9.80366307539372e-06,
+      "loss": 0.0052,
+      "step": 172
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 9.705518267520369e-06,
+      "loss": 0.0105,
+      "step": 173
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 9.607401842409318e-06,
+      "loss": 0.0089,
+      "step": 174
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 9.50932325672582e-06,
+      "loss": 0.0071,
+      "step": 175
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 9.41129196348811e-06,
+      "loss": 0.0082,
+      "step": 176
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 9.313317411156265e-06,
+      "loss": 0.0122,
+      "step": 177
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 9.215409042721553e-06,
+      "loss": 0.0077,
+      "step": 178
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 9.117576294796307e-06,
+      "loss": 0.0084,
+      "step": 179
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 9.019828596704394e-06,
+      "loss": 0.0099,
+      "step": 180
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 8.922175369572407e-06,
+      "loss": 0.0085,
+      "step": 181
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 8.824626025421625e-06,
+      "loss": 0.005,
+      "step": 182
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 8.72718996626087e-06,
+      "loss": 0.0068,
+      "step": 183
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 8.629876583180322e-06,
+      "loss": 0.0124,
+      "step": 184
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 8.532695255446384e-06,
+      "loss": 0.0093,
+      "step": 185
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 8.43565534959769e-06,
+      "loss": 0.0175,
+      "step": 186
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 8.338766218542348e-06,
+      "loss": 0.0053,
+      "step": 187
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 8.242037200656455e-06,
+      "loss": 0.0145,
+      "step": 188
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 8.145477618884092e-06,
+      "loss": 0.0096,
+      "step": 189
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 8.04909677983872e-06,
+      "loss": 0.0076,
+      "step": 190
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 7.952903972906205e-06,
+      "loss": 0.0092,
+      "step": 191
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 7.856908469349495e-06,
+      "loss": 0.0113,
+      "step": 192
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 7.761119521415017e-06,
+      "loss": 0.007,
+      "step": 193
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 7.66554636144095e-06,
+      "loss": 0.0144,
+      "step": 194
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 7.570198200967363e-06,
+      "loss": 0.0088,
+      "step": 195
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 7.4750842298484205e-06,
+      "loss": 0.0063,
+      "step": 196
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 7.380213615366627e-06,
+      "loss": 0.0083,
+      "step": 197
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 7.285595501349259e-06,
+      "loss": 0.0095,
+      "step": 198
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 7.191239007287082e-06,
+      "loss": 0.0042,
+      "step": 199
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 7.097153227455379e-06,
+      "loss": 0.0081,
+      "step": 200
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 7.003347230037434e-06,
+      "loss": 0.0102,
+      "step": 201
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 6.909830056250527e-06,
+      "loss": 0.0049,
+      "step": 202
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 6.816610719474503e-06,
+      "loss": 0.0084,
+      "step": 203
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 6.723698204383067e-06,
+      "loss": 0.0091,
+      "step": 204
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 6.631101466077801e-06,
+      "loss": 0.0084,
+      "step": 205
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 6.538829429225068e-06,
+      "loss": 0.0124,
+      "step": 206
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 6.446890987195842e-06,
+      "loss": 0.0051,
+      "step": 207
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 6.355295001208504e-06,
+      "loss": 0.0047,
+      "step": 208
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 6.2640502994748375e-06,
+      "loss": 0.0135,
+      "step": 209
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 6.173165676349103e-06,
+      "loss": 0.0095,
+      "step": 210
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 6.082649891480441e-06,
+      "loss": 0.0051,
+      "step": 211
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 5.9925116689685925e-06,
+      "loss": 0.0076,
+      "step": 212
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 5.902759696523046e-06,
+      "loss": 0.0051,
+      "step": 213
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 5.813402624625722e-06,
+      "loss": 0.0066,
+      "step": 214
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 5.724449065697182e-06,
+      "loss": 0.0097,
+      "step": 215
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 5.635907593266578e-06,
+      "loss": 0.0094,
+      "step": 216
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 5.54778674114532e-06,
+      "loss": 0.0093,
+      "step": 217
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 5.460095002604533e-06,
+      "loss": 0.0089,
+      "step": 218
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 5.3728408295565e-06,
+      "loss": 0.0063,
+      "step": 219
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 5.286032631740023e-06,
+      "loss": 0.0063,
+      "step": 220
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 5.199678775909889e-06,
+      "loss": 0.0052,
+      "step": 221
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 5.1137875850304545e-06,
+      "loss": 0.007,
+      "step": 222
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 5.0283673374734546e-06,
+      "loss": 0.0058,
+      "step": 223
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 4.943426266220156e-06,
+      "loss": 0.0046,
+      "step": 224
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 4.858972558067784e-06,
+      "loss": 0.0056,
+      "step": 225
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 4.775014352840512e-06,
+      "loss": 0.0092,
+      "step": 226
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 4.691559742604906e-06,
+      "loss": 0.0068,
+      "step": 227
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 4.608616770889998e-06,
+      "loss": 0.0071,
+      "step": 228
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 4.526193431912038e-06,
+      "loss": 0.0052,
+      "step": 229
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 4.444297669803981e-06,
+      "loss": 0.0092,
+      "step": 230
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 4.362937377849832e-06,
+      "loss": 0.0091,
+      "step": 231
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 4.282120397723879e-06,
+      "loss": 0.0053,
+      "step": 232
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 4.2018545187348645e-06,
+      "loss": 0.0068,
+      "step": 233
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 4.12214747707527e-06,
+      "loss": 0.0049,
+      "step": 234
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 4.043006955075667e-06,
+      "loss": 0.0046,
+      "step": 235
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 3.964440580464286e-06,
+      "loss": 0.0084,
+      "step": 236
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 3.8864559256318375e-06,
+      "loss": 0.0073,
+      "step": 237
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 3.8090605069016596e-06,
+      "loss": 0.0057,
+      "step": 238
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 3.7322617838053066e-06,
+      "loss": 0.0069,
+      "step": 239
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 3.6560671583635467e-06,
+      "loss": 0.0039,
+      "step": 240
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 3.58048397437297e-06,
+      "loss": 0.0044,
+      "step": 241
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 3.505519516698165e-06,
+      "loss": 0.0095,
+      "step": 242
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 3.4311810105695875e-06,
+      "loss": 0.0046,
+      "step": 243
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 3.3574756208871862e-06,
+      "loss": 0.0061,
+      "step": 244
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 3.284410451529816e-06,
+      "loss": 0.0041,
+      "step": 245
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 3.2119925446705824e-06,
+      "loss": 0.0065,
+      "step": 246
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 3.140228880098074e-06,
+      "loss": 0.0044,
+      "step": 247
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 3.069126374543643e-06,
+      "loss": 0.0046,
+      "step": 248
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 2.998691881014765e-06,
+      "loss": 0.0072,
+      "step": 249
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 2.9289321881345257e-06,
+      "loss": 0.0054,
+      "step": 250
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 2.859854019487318e-06,
+      "loss": 0.0087,
+      "step": 251
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 2.791464032970812e-06,
+      "loss": 0.0049,
+      "step": 252
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 2.723768820154251e-06,
+      "loss": 0.0065,
+      "step": 253
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 2.656774905643147e-06,
+      "loss": 0.0061,
+      "step": 254
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 2.5904887464504115e-06,
+      "loss": 0.0047,
+      "step": 255
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 2.5249167313740307e-06,
+      "loss": 0.0048,
+      "step": 256
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 2.4600651803813057e-06,
+      "loss": 0.0046,
+      "step": 257
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 2.395940343999691e-06,
+      "loss": 0.0057,
+      "step": 258
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 2.332548402714385e-06,
+      "loss": 0.0056,
+      "step": 259
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 2.26989546637263e-06,
+      "loss": 0.0124,
+      "step": 260
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 2.207987573594833e-06,
+      "loss": 0.0026,
+      "step": 261
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 2.146830691192553e-06,
+      "loss": 0.0098,
+      "step": 262
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 2.086430713593397e-06,
+      "loss": 0.0052,
+      "step": 263
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 2.02679346227293e-06,
+      "loss": 0.0053,
+      "step": 264
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 1.967924685193552e-06,
+      "loss": 0.0072,
+      "step": 265
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 1.9098300562505266e-06,
+      "loss": 0.0063,
+      "step": 266
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 1.8525151747251058e-06,
+      "loss": 0.0068,
+      "step": 267
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 1.7959855647448642e-06,
+      "loss": 0.0064,
+      "step": 268
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 1.7402466747512704e-06,
+      "loss": 0.006,
+      "step": 269
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 1.6853038769745466e-06,
+      "loss": 0.0048,
+      "step": 270
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 1.6311624669159064e-06,
+      "loss": 0.007,
+      "step": 271
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 1.577827662837136e-06,
+      "loss": 0.0042,
+      "step": 272
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 1.5253046052576559e-06,
+      "loss": 0.0044,
+      "step": 273
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 1.4735983564590784e-06,
+      "loss": 0.0049,
+      "step": 274
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 1.4227138999972801e-06,
+      "loss": 0.0058,
+      "step": 275
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 1.3726561402220818e-06,
+      "loss": 0.0062,
+      "step": 276
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 1.3234299018045615e-06,
+      "loss": 0.0066,
+      "step": 277
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 1.2750399292720284e-06,
+      "loss": 0.0064,
+      "step": 278
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 1.2274908865507595e-06,
+      "loss": 0.0052,
+      "step": 279
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 1.1807873565164507e-06,
+      "loss": 0.0077,
+      "step": 280
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 1.1349338405525368e-06,
+      "loss": 0.0036,
+      "step": 281
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 1.0899347581163222e-06,
+      "loss": 0.0076,
+      "step": 282
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 1.045794446313031e-06,
+      "loss": 0.0047,
+      "step": 283
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 1.0025171594777872e-06,
+      "loss": 0.0053,
+      "step": 284
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 9.601070687655667e-07,
+      "loss": 0.0049,
+      "step": 285
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 9.185682617491865e-07,
+      "loss": 0.0073,
+      "step": 286
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 8.779047420253239e-07,
+      "loss": 0.0056,
+      "step": 287
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 8.381204288286415e-07,
+      "loss": 0.0087,
+      "step": 288
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 7.992191566540519e-07,
+      "loss": 0.0041,
+      "step": 289
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 7.612046748871327e-07,
+      "loss": 0.0062,
+      "step": 290
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 7.240806474427598e-07,
+      "loss": 0.0035,
+      "step": 291
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 6.878506524119644e-07,
+      "loss": 0.0042,
+      "step": 292
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 6.525181817170756e-07,
+      "loss": 0.0063,
+      "step": 293
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 6.180866407751595e-07,
+      "loss": 0.0054,
+      "step": 294
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 5.845593481697931e-07,
+      "loss": 0.0053,
+      "step": 295
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 5.519395353312195e-07,
+      "loss": 0.008,
+      "step": 296
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 5.20230346224897e-07,
+      "loss": 0.0039,
+      "step": 297
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 4.894348370484648e-07,
+      "loss": 0.0037,
+      "step": 298
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 4.5955597593719593e-07,
+      "loss": 0.0061,
+      "step": 299
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 4.305966426779118e-07,
+      "loss": 0.0059,
+      "step": 300
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 4.025596284314259e-07,
+      "loss": 0.0067,
+      "step": 301
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 3.7544763546352834e-07,
+      "loss": 0.0064,
+      "step": 302
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 3.492632768845261e-07,
+      "loss": 0.0047,
+      "step": 303
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 3.2400907639740243e-07,
+      "loss": 0.0033,
+      "step": 304
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 2.996874680545603e-07,
+      "loss": 0.0081,
+      "step": 305
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 2.7630079602323447e-07,
+      "loss": 0.0065,
+      "step": 306
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 2.5385131435955e-07,
+      "loss": 0.005,
+      "step": 307
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 2.3234118679127615e-07,
+      "loss": 0.0051,
+      "step": 308
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 2.117724865092774e-07,
+      "loss": 0.0061,
+      "step": 309
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 1.921471959676957e-07,
+      "loss": 0.0076,
+      "step": 310
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 1.734672066928822e-07,
+      "loss": 0.0063,
+      "step": 311
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 1.5573431910108404e-07,
+      "loss": 0.0031,
+      "step": 312
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 1.3895024232491338e-07,
+      "loss": 0.0039,
+      "step": 313
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 1.231165940486234e-07,
+      "loss": 0.0053,
+      "step": 314
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 1.0823490035218986e-07,
+      "loss": 0.0078,
+      "step": 315
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 9.43065955642275e-08,
+      "loss": 0.0098,
+      "step": 316
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 8.133302212373961e-08,
+      "loss": 0.0054,
+      "step": 317
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 6.931543045073708e-08,
+      "loss": 0.0054,
+      "step": 318
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 5.8254978825718065e-08,
+      "loss": 0.0046,
+      "step": 319
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 4.815273327803183e-08,
+      "loss": 0.0086,
+      "step": 320
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 3.900966748312862e-08,
+      "loss": 0.0057,
+      "step": 321
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 3.082666266872036e-08,
+      "loss": 0.0055,
+      "step": 322
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 2.3604507529843e-08,
+      "loss": 0.0054,
+      "step": 323
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 1.7343898152841765e-08,
+      "loss": 0.004,
+      "step": 324
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 1.2045437948275952e-08,
+      "loss": 0.0031,
+      "step": 325
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 7.70963759277099e-09,
+      "loss": 0.0075,
+      "step": 326
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 4.336914979787832e-09,
+      "loss": 0.0055,
+      "step": 327
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 1.9275951793518154e-09,
+      "loss": 0.0054,
+      "step": 328
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 4.819104067199653e-10,
+      "loss": 0.0052,
+      "step": 329
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 0.0,
+      "loss": 0.0021,
+      "step": 330
+    },
+    {
+      "epoch": 3.0,
+      "step": 330,
+      "total_flos": 0.0,
+      "train_loss": 0.016219358733206086,
+      "train_runtime": 8423.2988,
+      "train_samples_per_second": 1.247,
+      "train_steps_per_second": 0.039
+    }
+  ],
+  "logging_steps": 1.0,
+  "max_steps": 330,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "total_flos": 0.0,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94099be84e09b0bf6180c33929e84583fce28f561bfc0c9e87ec3a0bfb39ad28
+size 6779

visual_encoder.py

@@ -0,0 +1,1019 @@
+import math
+from typing import Any, Optional, Tuple, Union
+
+from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling, BaseModelOutputWithPastAndCrossAttentions
+from transformers.modeling_utils import PreTrainedModel
+from transformers.pytorch_utils import find_pruneable_heads_and_indices, prune_linear_layer
+
+import numpy as np
+import torch
+import torch.nn as nn
+import torch.utils.checkpoint
+# icecream removed for inference
+
+def get_abs_pos(abs_pos, tgt_size):
+    # abs_pos: L, C
+    # tgt_size: M
+    # return: M, C
+    src_size = int(math.sqrt(abs_pos.size(0)))
+    tgt_size = int(math.sqrt(tgt_size))
+    dtype = abs_pos.dtype
+
+    if src_size != tgt_size:
+        return F.interpolate(
+            abs_pos.float().reshape(1, src_size, src_size, -1).permute(0, 3, 1, 2),
+            size=(tgt_size, tgt_size),
+            mode="bicubic",
+            align_corners=False,
+        ).permute(0, 2, 3, 1).flatten(0, 2).to(dtype=dtype)
+    else:
+        return abs_pos
+
+# https://github.com/facebookresearch/mae/blob/efb2a8062c206524e35e47d04501ed4f544c0ae8/util/pos_embed.py#L20
+def get_2d_sincos_pos_embed(embed_dim, grid_size, cls_token=False):
+    """
+    grid_size: int of the grid height and width
+    return:
+    pos_embed: [grid_size*grid_size, embed_dim] or [1+grid_size*grid_size, embed_dim] (w/ or w/o cls_token)
+    """
+    grid_h = np.arange(grid_size, dtype=np.float32)
+    grid_w = np.arange(grid_size, dtype=np.float32)
+    grid = np.meshgrid(grid_w, grid_h)  # here w goes first
+    grid = np.stack(grid, axis=0)
+
+    grid = grid.reshape([2, 1, grid_size, grid_size])
+    pos_embed = get_2d_sincos_pos_embed_from_grid(embed_dim, grid)
+    if cls_token:
+        pos_embed = np.concatenate([np.zeros([1, embed_dim]), pos_embed], axis=0)
+    return pos_embed
+
+
+def get_2d_sincos_pos_embed_from_grid(embed_dim, grid):
+    assert embed_dim % 2 == 0
+
+    # use half of dimensions to encode grid_h
+    emb_h = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[0])  # (H*W, D/2)
+    emb_w = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[1])  # (H*W, D/2)
+
+    emb = np.concatenate([emb_h, emb_w], axis=1) # (H*W, D)
+    return emb
+
+
+def get_1d_sincos_pos_embed_from_grid(embed_dim, pos):
+    """
+    embed_dim: output dimension for each position
+    pos: a list of positions to be encoded: size (M,)
+    out: (M, D)
+    """
+    assert embed_dim % 2 == 0
+    omega = np.arange(embed_dim // 2, dtype=np.float32)
+    omega /= embed_dim / 2.
+    omega = 1. / 10000**omega  # (D/2,)
+
+    pos = pos.reshape(-1)  # (M,)
+    out = np.einsum('m,d->md', pos, omega)  # (M, D/2), outer product
+
+    emb_sin = np.sin(out) # (M, D/2)
+    emb_cos = np.cos(out) # (M, D/2)
+
+    emb = np.concatenate([emb_sin, emb_cos], axis=1)  # (M, D)
+    return emb
+
+
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+class MplugOwlVisionEmbeddings(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.image_size = config.image_size
+        self.patch_size = config.patch_size
+
+        self.cls_token = nn.Parameter(torch.randn(1, 1, self.hidden_size))
+
+        self.patch_embed = nn.Conv2d(
+            in_channels=3,
+            out_channels=self.hidden_size,
+            kernel_size=self.patch_size,
+            stride=self.patch_size,
+            bias=False,
+        )
+
+        # Initialize position embedding for default size (can be resized later)
+        self.num_patches = (self.image_size // self.patch_size) ** 2
+        self.position_embedding = nn.Parameter(torch.randn(1, self.num_patches + 1, self.hidden_size))
+        self.pre_layernorm = nn.LayerNorm(self.hidden_size, eps=config.layer_norm_eps)
+
+    def interpolate_pos_encoding(self, embeddings, h, w):
+        """
+        Interpolate position embeddings for different image sizes
+        """
+        npatch = embeddings.shape[1] - 1  # subtract 1 for cls token
+        N = self.position_embedding.shape[1] - 1  # original number of patches
+        
+        if npatch == N:
+            return self.position_embedding
+        
+        # Separate class token and patch embeddings
+        class_pos_embed = self.position_embedding[:, 0:1]  # [1, 1, hidden_size]
+        patch_pos_embed = self.position_embedding[:, 1:]   # [1, N, hidden_size]
+        
+        dim = embeddings.shape[-1]
+        
+        # Calculate original grid size
+        w0 = h0 = int(N ** 0.5)
+        
+        # Reshape patch embeddings to 2D grid
+        patch_pos_embed = patch_pos_embed.reshape(1, w0, h0, dim).permute(0, 3, 1, 2)
+        
+        # Convert to float32 for interpolation
+        patch_pos_embed = patch_pos_embed.float()
+        
+        # Interpolate to new size
+        patch_pos_embed = F.interpolate(
+            patch_pos_embed,
+            size=(h, w),
+            mode='bicubic',
+            align_corners=False,
+        )
+        
+        # Convert back to original dtype
+        patch_pos_embed = patch_pos_embed.to(dtype=embeddings.dtype)
+        
+        # Reshape back to sequence
+        patch_pos_embed = patch_pos_embed.permute(0, 2, 3, 1).reshape(1, -1, dim)
+        
+        # Concatenate class token and patch embeddings
+        return torch.cat((class_pos_embed, patch_pos_embed), dim=1)
+
+    def forward(self, pixel_values: torch.FloatTensor) -> torch.Tensor:
+        batch_size = pixel_values.size(0)
+        #print(f"[DEBUG] Input image shape: {pixel_values.shape}")
+        
+        image_embeds = self.patch_embed(pixel_values)
+        #print(f"[DEBUG] After patch_embed shape: {image_embeds.shape}")
+        
+        # Get patch grid dimensions
+        _, _, h, w = image_embeds.shape
+        
+        image_embeds = image_embeds.flatten(2).transpose(1, 2)
+        #print(f"[DEBUG] After flatten and transpose shape: {image_embeds.shape}")
+        
+        class_embeds = self.cls_token.expand(batch_size, 1, -1).to(image_embeds.dtype)
+        embeddings = torch.cat([class_embeds, image_embeds], dim=1)
+        
+        # Interpolate position embeddings to match current image size
+        pos_embed = self.interpolate_pos_encoding(embeddings, h, w).to(image_embeds.dtype)
+        #print(f"[DEBUG] Position embedding shape after interpolation: {pos_embed.shape}")
+        
+        embeddings = embeddings + pos_embed
+        embeddings = self.pre_layernorm(embeddings)
+        return embeddings
+
+
+
+class MplugOwlVisionAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.hidden_size // self.num_heads
+        if self.head_dim * self.num_heads != self.hidden_size:
+            raise ValueError(
+                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size} and `num_heads`:"
+                f" {self.num_heads})."
+            )
+        self.scale = self.head_dim**-0.5
+        self.dropout = nn.Dropout(config.attention_dropout)
+
+        self.query_key_value = nn.Linear(self.hidden_size, 3 * self.hidden_size)
+        self.dense = nn.Linear(self.hidden_size, self.hidden_size)
+
+    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        head_mask: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        """Input shape: Batch x Time x Channel"""
+
+        bsz, seq_len, embed_dim = hidden_states.size()
+
+        mixed_qkv = self.query_key_value(hidden_states)
+
+        mixed_qkv = mixed_qkv.reshape(bsz, seq_len, self.num_heads, 3, embed_dim // self.num_heads).permute(
+            3, 0, 2, 1, 4
+        )  # [3, b, np, sq, hn]
+        query_states, key_states, value_states = (
+            mixed_qkv[0],
+            mixed_qkv[1],
+            mixed_qkv[2],
+        )
+        # if self.config.use_flash_attn and flash_attn_func is not None:
+        if False:
+            # [b*sq, np, hn]
+            query_states = query_states.permute(0, 2, 1, 3).contiguous()
+            query_states = query_states.view(query_states.size(0) * query_states.size(1), query_states.size(2), -1)
+
+            key_states = key_states.permute(0, 2, 1, 3).contiguous()
+            key_states = key_states.view(key_states.size(0) * key_states.size(1), key_states.size(2), -1)
+
+            value_states = value_states.permute(0, 2, 1, 3).contiguous()
+            value_states = value_states.view(value_states.size(0) * value_states.size(1), value_states.size(2), -1)
+
+            cu_seqlens = torch.arange(
+                0, (bsz + 1) * seq_len, step=seq_len, dtype=torch.int32, device=query_states.device
+            )
+
+            context_layer = flash_attn_func(
+                query_states,
+                key_states,
+                value_states,
+                cu_seqlens,
+                cu_seqlens,
+                seq_len,
+                seq_len,
+                self.dropout if self.training else 0.0,
+                softmax_scale=self.scale,
+                causal=False,
+                return_attn_probs=False,
+            )
+            # [b*sq, np, hn] => [b, sq, np, hn]
+            context_layer = context_layer.view(bsz, seq_len, context_layer.size(1), context_layer.size(2))
+        else:
+            # Take the dot product between "query" and "key" to get the raw attention scores.
+            attention_scores = torch.matmul(query_states, key_states.transpose(-1, -2))
+
+            attention_scores = attention_scores * self.scale
+
+            # Normalize the attention scores to probabilities.
+            attention_probs = torch.softmax(attention_scores, dim=-1)
+
+            # This is actually dropping out entire tokens to attend to, which might
+            # seem a bit unusual, but is taken from the original Transformer paper.
+            attention_probs = self.dropout(attention_probs)
+
+            # Mask heads if we want to
+            if head_mask is not None:
+                attention_probs = attention_probs * head_mask
+
+            context_layer = torch.matmul(attention_probs, value_states).permute(0, 2, 1, 3)
+
+        new_context_layer_shape = context_layer.size()[:-2] + (self.hidden_size,)
+        context_layer = context_layer.reshape(new_context_layer_shape)
+
+        output = self.dense(context_layer)
+
+        outputs = (output, attention_probs) if output_attentions else (output, None)
+
+        return outputs
+
+
+class QuickGELU(nn.Module):
+    def forward(self, x: torch.Tensor):
+        return x * torch.sigmoid(1.702 * x)
+
+
+class MplugOwlMLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.activation_fn = QuickGELU()
+        self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
+        self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.fc1(hidden_states)
+        hidden_states = self.activation_fn(hidden_states)
+        hidden_states = self.fc2(hidden_states)
+        return hidden_states
+
+
+class MplugOwlVisionEncoderLayer(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.self_attn = MplugOwlVisionAttention(config)
+        self.input_layernorm = nn.LayerNorm(self.hidden_size, eps=config.layer_norm_eps)
+        self.mlp = MplugOwlMLP(config)
+        self.post_attention_layernorm = nn.LayerNorm(self.hidden_size, eps=config.layer_norm_eps)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: torch.Tensor,
+        output_attentions: Optional[bool] = False,
+    ) -> Tuple[torch.FloatTensor]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`): attention mask of size
+                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
+                `(config.encoder_attention_heads,)`.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+        """
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+        hidden_states, attn_weights = self.self_attn(
+            hidden_states=hidden_states,
+            head_mask=attention_mask,
+            output_attentions=output_attentions,
+        )
+        hidden_states = hidden_states + residual
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+
+        hidden_states = hidden_states + residual
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (attn_weights,)
+
+        return outputs
+    
+    
+class MplugOwlVisionEncoder(nn.Module):
+    """
+    Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a
+    [`MplugOwlVisionEncoderLayer`].
+
+    Args:
+        config (`MplugOwlVisionConfig`):
+            The corresponding vision configuration for the `MplugOwlEncoder`.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.layers = nn.ModuleList([MplugOwlVisionEncoderLayer(config) for _ in range(config.num_hidden_layers)])
+        self.gradient_checkpointing = True
+
+    def forward(
+        self,
+        inputs_embeds,
+        attention_mask: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutput]:
+        r"""
+        Args:
+            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
+                Embedded representation of the inputs. Should be float, not int tokens.
+            attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+                - 1 for tokens that are **not masked**,
+                - 0 for tokens that are **masked**.
+
+                [What are attention masks?](../glossary#attention-mask)
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            output_hidden_states (`bool`, *optional*):
+                Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
+                for more detail.
+            return_dict (`bool`, *optional*):
+                Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+        """
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        encoder_states = () if output_hidden_states else None
+        all_attentions = () if output_attentions else None
+
+        hidden_states = inputs_embeds
+        for idx, encoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                encoder_states = encoder_states + (hidden_states,)
+            if self.gradient_checkpointing and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        return module(*inputs, output_attentions)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(encoder_layer),
+                    hidden_states,
+                    attention_mask,
+                )
+            else:
+                layer_outputs = encoder_layer(
+                    hidden_states,
+                    attention_mask,
+                    output_attentions=output_attentions,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if output_attentions:
+                all_attentions = all_attentions + (layer_outputs[1],)
+
+        if output_hidden_states:
+            encoder_states = encoder_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(v for v in [hidden_states, encoder_states, all_attentions] if v is not None)
+        return BaseModelOutput(
+            last_hidden_state=hidden_states, hidden_states=encoder_states, attentions=all_attentions
+        )
+
+
+class MplugOwlVisionModel(PreTrainedModel):
+    main_input_name = "pixel_values"
+    _no_split_modules = ["MplugOwlVisionEncoderLayer"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.config = config
+        self.hidden_size = config.hidden_size
+
+        self.embeddings = MplugOwlVisionEmbeddings(config)
+        self.encoder = MplugOwlVisionEncoder(config)
+        self.post_layernorm = nn.LayerNorm(self.hidden_size, eps=config.layer_norm_eps)
+
+        self.post_init()
+
+
+    def forward(
+        self,
+        pixel_values: Optional[torch.FloatTensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPooling]:
+        r"""
+        Returns:
+
+        """
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if pixel_values is None:
+            raise ValueError("You have to specify pixel_values")
+
+        hidden_states = self.embeddings(pixel_values)
+
+        encoder_outputs = self.encoder(
+            inputs_embeds=hidden_states,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        last_hidden_state = encoder_outputs[0]
+        last_hidden_state = self.post_layernorm(last_hidden_state)
+
+        pooled_output = last_hidden_state[:, 0, :]
+        pooled_output = self.post_layernorm(pooled_output)
+
+        if not return_dict:
+            return (last_hidden_state, pooled_output) + encoder_outputs[1:]
+
+        return BaseModelOutputWithPooling(
+            last_hidden_state=last_hidden_state,
+            pooler_output=pooled_output,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+        )
+
+    def get_input_embeddings(self):
+        return self.embeddings
+
+
+class MplugOwlVisualAbstractorMLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        in_features = config.hidden_size
+        self.act = nn.SiLU()
+
+        self.w1 = nn.Linear(in_features, config.intermediate_size)
+        self.w2 = nn.Linear(config.intermediate_size, in_features)
+        self.w3 = nn.Linear(in_features, config.intermediate_size)
+        self.ffn_ln = nn.LayerNorm(config.intermediate_size, eps=config.layer_norm_eps)
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.act(self.w1(hidden_states)) * self.w3(hidden_states)
+        hidden_states = self.ffn_ln(hidden_states)
+        hidden_states = self.w2(hidden_states)
+        return hidden_states
+
+
+class MplugOwlVisualAbstractorMultiHeadAttention(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        if config.hidden_size % config.num_attention_heads != 0:
+            raise ValueError(
+                "The hidden size (%d) is not a multiple of the number of attention heads (%d)"
+                % (config.hidden_size, config.num_attention_heads)
+            )
+
+        self.num_attention_heads = config.num_attention_heads
+        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
+        self.all_head_size = self.num_attention_heads * self.attention_head_size
+
+        self.query = nn.Linear(config.hidden_size, self.all_head_size)
+        self.key = nn.Linear(config.encoder_hidden_size, self.all_head_size)
+        self.value = nn.Linear(config.encoder_hidden_size, self.all_head_size)
+
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+        self.save_attention = False
+        
+#         self.q_pos_embed = nn.Parameter(
+#             torch.from_numpy(get_1d_sincos_pos_embed_from_grid(config.hidden_size, np.arange(config.num_learnable_queries, dtype=np.float32))).float()
+#         ).requires_grad_(False)
+#         grids = config.grid_size
+#         self.k_pos_embed = nn.Parameter(
+#             torch.from_numpy(get_2d_sincos_pos_embed(config.hidden_size, grids, cls_token=True)).float()
+#         ).requires_grad_(False)
+        grids = config.grid_size
+        self.register_buffer(
+            'q_pos_embed', 
+            torch.from_numpy(get_1d_sincos_pos_embed_from_grid(config.hidden_size, np.arange(config.num_learnable_queries, dtype=np.float32))).float()
+        )
+        self.register_buffer(
+            'k_pos_embed', 
+            torch.from_numpy(get_2d_sincos_pos_embed(config.hidden_size, grids, cls_token=True)).float()
+        )
+        
+
+    def save_attn_gradients(self, attn_gradients):
+        self.attn_gradients = attn_gradients
+
+    def get_attn_gradients(self):
+        return self.attn_gradients
+
+    def save_attention_map(self, attention_map):
+        self.attention_map = attention_map
+
+    def get_attention_map(self):
+        return self.attention_map
+
+    def transpose_for_scores(self, x):
+        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
+        x = x.view(*new_x_shape)
+        return x.permute(0, 2, 1, 3)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):
+        # If this is instantiated as a cross-attention module, the keys
+        # and values come from an encoder; the attention mask needs to be
+        # such that the encoder's padding tokens are not attended to.
+        
+        # 确保位置编码的维度与输入匹配
+        if encoder_hidden_states is not None:
+            seq_len = encoder_hidden_states.size(1)
+            if seq_len != self.k_pos_embed.size(0):
+                # 如果序列长度不匹配，需要调整位置编码
+                # 使用更高效的方式调整位置编码
+                k_pos_embed = self.k_pos_embed
+                if seq_len > k_pos_embed.size(0):
+                    # 如果目标序列更长，使用重复
+                    repeat_times = (seq_len + k_pos_embed.size(0) - 1) // k_pos_embed.size(0)
+                    k_pos_embed = k_pos_embed.repeat(repeat_times, 1)[:seq_len]
+                else:
+                    # 如果目标序列更短，使用切片
+                    k_pos_embed = k_pos_embed[:seq_len]
+            else:
+                k_pos_embed = self.k_pos_embed
+                
+            # 确保 q_pos_embed 和 k_pos_embed 的维度正确
+            q_pos_embed = self.q_pos_embed.to(dtype=hidden_states.dtype)
+            k_pos_embed = k_pos_embed.to(dtype=encoder_hidden_states.dtype)
+            
+            # 确保维度匹配
+            if q_pos_embed.size(0) + k_pos_embed.size(0) != encoder_hidden_states.size(1):
+                # 如果维度不匹配，调整 k_pos_embed 的大小
+                target_size = encoder_hidden_states.size(1) - q_pos_embed.size(0)
+                if target_size > k_pos_embed.size(0):
+                    # 如果目标大小更大，使用重复
+                    repeat_times = (target_size + k_pos_embed.size(0) - 1) // k_pos_embed.size(0)
+                    k_pos_embed = k_pos_embed.repeat(repeat_times, 1)[:target_size]
+                else:
+                    # 如果目标大小更小，使用切片
+                    k_pos_embed = k_pos_embed[:target_size]
+            
+            qk_pos_embed = torch.cat([q_pos_embed, k_pos_embed], dim=0).unsqueeze(0)
+        else:
+            qk_pos_embed = self.q_pos_embed.unsqueeze(0).to(dtype=hidden_states.dtype)
+        
+        # 确保最终维度匹配
+        assert qk_pos_embed.size(1) == encoder_hidden_states.size(1), \
+            f"Position embedding size {qk_pos_embed.size(1)} does not match encoder hidden states size {encoder_hidden_states.size(1)}"
+        
+        key_layer = self.transpose_for_scores(self.key(encoder_hidden_states + qk_pos_embed))
+        value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
+        attention_mask = encoder_attention_mask
+
+        mixed_query_layer = self.query(hidden_states + self.q_pos_embed.unsqueeze(0).to(dtype=hidden_states.dtype))
+
+        query_layer = self.transpose_for_scores(mixed_query_layer)
+
+        past_key_value = (key_layer, value_layer)
+
+        # Take the dot product between "query" and "key" to get the raw attention scores.
+        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
+
+        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
+
+        if attention_mask is not None:
+            # Apply the attention mask is (precomputed for all layers in BertModel forward() function)
+            attention_scores = attention_scores + attention_mask
+
+        # Normalize the attention scores to probabilities.
+        attention_probs = nn.Softmax(dim=-1)(attention_scores)
+
+        if self.save_attention:
+            self.save_attention_map(attention_probs)
+            attention_probs.register_hook(self.save_attn_gradients)
+
+        # This is actually dropping out entire tokens to attend to, which might
+        # seem a bit unusual, but is taken from the original Transformer paper.
+        attention_probs_dropped = self.dropout(attention_probs)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            attention_probs_dropped = attention_probs_dropped * head_mask
+
+        context_layer = torch.matmul(attention_probs_dropped, value_layer)
+
+        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
+        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
+        context_layer = context_layer.view(*new_context_layer_shape)
+
+        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)
+
+        outputs = outputs + (past_key_value,)
+        return outputs
+
+
+class MplugOwlVisualAbstractorCrossOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        dim = config.hidden_size
+        self.out_proj = nn.Linear(dim, dim, bias=True)
+        self.norm2 = nn.LayerNorm(dim)
+        self.mlp = MplugOwlVisualAbstractorMLP(config)
+
+    def forward(self, hidden_states: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor:
+        input_tensor = input_tensor + self.out_proj(hidden_states)
+        input_tensor = input_tensor + self.mlp(self.norm2(input_tensor))
+        return input_tensor
+
+
+class MplugOwlVisualAbstractorAttention(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.attention = MplugOwlVisualAbstractorMultiHeadAttention(config)
+        self.output = MplugOwlVisualAbstractorCrossOutput(config)
+        self.pruned_heads = set()
+        self.norm1 = nn.LayerNorm(config.hidden_size)
+        self.normk = nn.LayerNorm(config.hidden_size)
+
+    def prune_heads(self, heads):
+        if len(heads) == 0:
+            return
+        heads, index = find_pruneable_heads_and_indices(
+            heads, self.attention.num_attention_heads, self.attention.attention_head_size, self.pruned_heads
+        )
+
+        # Prune linear layers
+        self.attention.query = prune_linear_layer(self.attention.query, index)
+        self.attention.key = prune_linear_layer(self.attention.key, index)
+        self.attention.value = prune_linear_layer(self.attention.value, index)
+        self.output.dense = prune_linear_layer(self.output.out_proj, index, dim=1)
+
+        # Update hyper params and store pruned heads
+        self.attention.num_attention_heads = self.attention.num_attention_heads - len(heads)
+        self.attention.all_head_size = self.attention.attention_head_size * self.attention.num_attention_heads
+        self.pruned_heads = self.pruned_heads.union(heads)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        encoder_hidden_states: Optional[torch.FloatTensor] = None,
+        encoder_attention_mask: Optional[torch.FloatTensor] = None,
+        past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
+        output_attentions: Optional[bool] = False,
+    ) -> Tuple[torch.Tensor]:
+        # HACK we apply norm on q and k
+        hidden_states = self.norm1(hidden_states)
+        encoder_hidden_states = self.normk(encoder_hidden_states)
+        encoder_hidden_states = torch.cat([hidden_states, encoder_hidden_states], dim=1)
+        encoder_attention_mask = torch.cat([attention_mask, encoder_attention_mask], dim=-1)
+        self_outputs = self.attention(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            encoder_hidden_states,
+            encoder_attention_mask,
+            past_key_value,
+            output_attentions,
+        )
+        attention_output = self.output(self_outputs[0], hidden_states)
+        # add attentions if we output them
+        outputs = (attention_output,) + self_outputs[1:]
+        return outputs
+
+
+class MplugOwlVisualAbstractorLayer(nn.Module):
+    def __init__(self, config, layer_idx):
+        super().__init__()
+        self.chunk_size_feed_forward = config.chunk_size_feed_forward
+        self.seq_len_dim = 1
+
+        self.layer_idx = layer_idx
+
+        self.crossattention = MplugOwlVisualAbstractorAttention(config)
+        self.has_cross_attention = True
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        output_attentions=False,
+    ):
+        if encoder_hidden_states is None:
+            raise ValueError("encoder_hidden_states must be given for cross-attention layers")
+        cross_attention_outputs = self.crossattention(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            encoder_hidden_states,
+            encoder_attention_mask,
+            output_attentions=output_attentions,
+        )
+        query_attention_output = cross_attention_outputs[0]
+
+        outputs = (query_attention_output,)
+        return outputs
+
+
+class MplugOwlVisualAbstractorEncoder(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.layers = nn.ModuleList(
+            [MplugOwlVisualAbstractorLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.gradient_checkpointing = True
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        output_attentions=False,
+        output_hidden_states=False,
+        return_dict=True,
+    ):
+        all_hidden_states = () if output_hidden_states else None
+
+        for i in range(self.config.num_hidden_layers):
+            layer_module = self.layers[i]
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            layer_head_mask = head_mask[i] if head_mask is not None else None
+            past_key_value = past_key_values[i] if past_key_values is not None else None
+
+            if getattr(self.config, "gradient_checkpointing", False) and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        return module(*inputs, past_key_value, output_attentions)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(layer_module),
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                )
+            else:
+                layer_outputs = layer_module(
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    output_attentions,
+                )
+
+            hidden_states = layer_outputs[0]
+
+        return BaseModelOutput(
+            last_hidden_state=hidden_states,
+        )
+
+
+class MplugOwlVisualAbstractorModel(PreTrainedModel):
+    _no_split_modules = ["MplugOwlVisualAbstractorLayer"]
+    def __init__(self, config, language_hidden_size):
+        super().__init__(config)
+        self.config = config
+
+        self.encoder = MplugOwlVisualAbstractorEncoder(config)
+        self.visual_fc = torch.nn.Linear(config.hidden_size, language_hidden_size)
+        self.query_embeds = torch.nn.Parameter(torch.randn(1, config.num_learnable_queries, config.hidden_size))
+        self.vit_eos = torch.nn.Parameter(torch.randn(1, 1, language_hidden_size))
+
+        self.post_init()
+
+    def _prune_heads(self, heads_to_prune):
+        """
+        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
+        class PreTrainedModel
+        """
+        for layer, heads in heads_to_prune.items():
+            self.encoder.layer[layer].attention.prune_heads(heads)
+
+    def get_extended_attention_mask(
+        self,
+        attention_mask: torch.Tensor,
+        input_shape: Tuple[int],
+        device: torch.device,
+    ) -> torch.Tensor:
+        """
+        Makes broadcastable attention and causal masks so that future and masked tokens are ignored.
+
+        Arguments:
+            attention_mask (`torch.Tensor`):
+                Mask with ones indicating tokens to attend to, zeros for tokens to ignore.
+            input_shape (`Tuple[int]`):
+                The shape of the input to the model.
+            device: (`torch.device`):
+                The device of the input to the model.
+
+        Returns:
+            `torch.Tensor` The extended attention mask, with a the same dtype as `attention_mask.dtype`.
+        """
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        if attention_mask.dim() == 3:
+            extended_attention_mask = attention_mask[:, None, :, :]
+        elif attention_mask.dim() == 2:
+            # Provided a padding mask of dimensions [batch_size, seq_length]
+            # - the model is an encoder, so make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
+            extended_attention_mask = attention_mask[:, None, None, :]
+        else:
+            raise ValueError(
+                "Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
+                    input_shape, attention_mask.shape
+                )
+            )
+
+        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
+        # masked positions, this operation will create a tensor which is 0.0 for
+        # positions we want to attend and -10000.0 for masked positions.
+        # Since we are adding it to the raw scores before the softmax, this is
+        # effectively the same as removing these entirely.
+        extended_attention_mask = extended_attention_mask.to(dtype=self.dtype)  # fp16 compatibility
+        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
+        return extended_attention_mask
+
+    def forward(
+        self,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        encoder_hidden_states  (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, `optional`):
+            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
+            the model is configured as a decoder.
+        encoder_attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, `optional`):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`:
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        past_key_values (`tuple(tuple(torch.FloatTensor))` of length `config.n_layers` with each tuple having 4 tensors of:
+            shape `(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`): Contains precomputed key and
+            value hidden states of the attention blocks. Can be used to speed up decoding. If `past_key_values` are
+            used, the user can optionally input only the last `decoder_input_ids` (those that don't have their past key
+            value states given to this model) of shape `(batch_size, 1)` instead of all `decoder_input_ids` of shape
+            `(batch_size, sequence_length)`.
+        """
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        
+        query_embeds = self.query_embeds.repeat(encoder_hidden_states.shape[0], 1, 1)
+        embedding_output = query_embeds
+        input_shape = embedding_output.size()[:-1]
+        batch_size, seq_length = input_shape
+        device = embedding_output.device
+
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        if attention_mask is None:
+            attention_mask = torch.ones(
+                (query_embeds.shape[0], query_embeds.shape[1]), dtype=torch.long, device=query_embeds.device
+            )
+        extended_attention_mask = self.get_extended_attention_mask(attention_mask, input_shape, device)
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if encoder_hidden_states is not None:
+            if type(encoder_hidden_states) == list:
+                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[0].size()
+            else:
+                (
+                    encoder_batch_size,
+                    encoder_sequence_length,
+                    _,
+                ) = encoder_hidden_states.size()
+            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
+
+            if type(encoder_attention_mask) == list:
+                encoder_extended_attention_mask = [self.invert_attention_mask(mask) for mask in encoder_attention_mask]
+            elif encoder_attention_mask is None:
+                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
+                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+            else:
+                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_extended_attention_mask = None
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
+        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
+        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
+
+        encoder_outputs = self.encoder(
+            embedding_output,
+            attention_mask=extended_attention_mask,
+            head_mask=head_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_extended_attention_mask,
+            past_key_values=past_key_values,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        sequence_output = encoder_outputs[0]
+        pooled_output = sequence_output[:, 0, :]
+
+        sequence_output = self.visual_fc(sequence_output)
+        sequence_output = torch.cat([sequence_output, self.vit_eos.repeat(sequence_output.shape[0], 1, 1)], dim=1)
+
+        return BaseModelOutputWithPooling(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            hidden_states=encoder_outputs.hidden_states,
+        )
+    
+