README.md

metadata

title: Forgekit
app_file: app.py
sdk: gradio
sdk_version: 5.42.0

🔥 ForgeKit

Forge your perfect AI model — no code required.

ForgeKit is an open-source platform that lets anyone create custom AI models by merging existing ones. No coding, no complex setup — just pick your models, configure the merge, and get a ready-to-run Colab notebook.

✨ Features

⚒️ Merge Builder

• Add models by ID and instantly check architecture compatibility
• Choose from 6 merge methods: DARE-TIES, TIES, SLERP, Linear, Task Arithmetic, Passthrough
• Adjust weights and densities with smart presets
• Auto-suggest base model and tokenizer
• Generate ready-to-run Colab notebooks with one click

🔍 Model Explorer

• Search HuggingFace Hub for models
• Filter by architecture type
• View detailed model specs (hidden size, layers, vocab, etc.)

📦 GGUF Quantizer

• Convert any HF model to GGUF format
• Multiple quantization levels (Q8_0, Q5_K_M, Q4_K_M, etc.)
• Ready-to-run Colab notebook generation

🚀 Deploy

• Generate deployment files for HuggingFace Spaces
• Gradio chat interface or Docker + llama.cpp options
• Auto-generated app.py and README

🏆 Community Leaderboard

• Browse community-created merges
• Submit your own merged models
• Discover popular merge recipes

🛠️ Supported Merge Methods

Method	Models	Best For
DARE-TIES	2-10	Combining specialists (coding + math)
TIES	2-10	Resolving parameter interference
SLERP	2	Smooth two-model interpolation
Linear	2-10	Simple weighted averaging
Task Arithmetic	1-10	Adding/removing capabilities
Passthrough	1-10	Layer stacking (Frankenmerge)

🚀 How It Works

1. Add Models — Enter HuggingFace model IDs
2. Check Compatibility — ForgeKit verifies architectures match
3. Configure — Choose method, adjust weights, pick presets
4. Generate — Get a Colab notebook with everything pre-filled
5. Run — Open in Colab, click Run All, wait for your model
6. Ship — Auto-upload to HF Hub + optional GGUF + Space deployment

📋 Requirements

The generated Colab notebooks handle all dependencies. You just need:

• A Google account (for Colab)
• A HuggingFace account (for model access and upload)
• A HF token (for gated models and uploading)

🧑‍💻 Built By

AIencoder — AI/ML Engineer

• Portfolio
• GitHub

📄 License

MIT — use it, fork it, improve it.