AIPrep/gem-early-ckpt
Image-Text-to-Text • 24B • Updated
• 45
AI & ML interests
Building GEM I Stay tuned
Recent Activity
KingNish
updated a model about 16 hours ago
Post
3123
Muon vs MuonClip vs Muon+Adamw
Muon has gone from an experiment to a mainstream optimizer, but does it hold up for fine‑tuning? We ran head‑to‑head tests on Qwen3‑4B (10k+ high‑quality instruction rows) to find out.
Short story: Pure Muon converged fastest at the start, but its gradient‑norm spikes made training unstable. MuonClip (Kimi K2’s clipping) stabilizes long pretraining runs, yet in our small‑scale fine‑tune it underperformed, lower token accuracy and slower convergence. The winner was the hybrid: Muon for 2D layers + AdamW for 1D layers. It delivered the best balance of stability and final performance and even beat vanilla AdamW.
Takeaway: for small-scale fine-tuning, hybrid = practical and reliable.
Next Step: scale to larger models/datasets to see if Muon’s spikes become catastrophic or if clipping wins out.
Full Blog Link: https://huggingface.co/blog/KingNish/optimizer-part1
Muon has gone from an experiment to a mainstream optimizer, but does it hold up for fine‑tuning? We ran head‑to‑head tests on Qwen3‑4B (10k+ high‑quality instruction rows) to find out.
Short story: Pure Muon converged fastest at the start, but its gradient‑norm spikes made training unstable. MuonClip (Kimi K2’s clipping) stabilizes long pretraining runs, yet in our small‑scale fine‑tune it underperformed, lower token accuracy and slower convergence. The winner was the hybrid: Muon for 2D layers + AdamW for 1D layers. It delivered the best balance of stability and final performance and even beat vanilla AdamW.
Takeaway: for small-scale fine-tuning, hybrid = practical and reliable.
Next Step: scale to larger models/datasets to see if Muon’s spikes become catastrophic or if clipping wins out.
Full Blog Link: https://huggingface.co/blog/KingNish/optimizer-part1
Post
2691
I tested Muon vs MuonClip vs Muon+AdamW for fine-tuning LLMs
Just published a blog on that, Read here 👉 https://huggingface.co/blog/KingNish/optimizer-part1
Just published a blog on that, Read here 👉 https://huggingface.co/blog/KingNish/optimizer-part1
- 1 reply
Post
2216
Wan 2.2 fast upto 10x faster than original wan 2.2
Model: FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers
Space: KingNish/wan2-2-fast
Model: FastVideo/FastWan2.2-TI2V-5B-FullAttn-Diffusers
Space: KingNish/wan2-2-fast
Post
1218
What's currently the biggest gap in Open Source Datasets ??
- 5 replies
Post
11075
Realtime Whisper Large v3 Turbo Demo:
It transcribes audio in about 0.3 seconds.
KingNish/Realtime-whisper-large-v3-turbo
It transcribes audio in about 0.3 seconds.
KingNish/Realtime-whisper-large-v3-turbo
- 2 replies
Post
8325
Exciting news! Introducing super-fast AI video assistant, currently in beta. With a minimum latency of under 500ms and an average latency of just 600ms.
DEMO LINK:
KingNish/Live-Video-Chat
DEMO LINK:
KingNish/Live-Video-Chat
- 1 reply
Post
4213
A super good and fast image inpainting demo is here.
Its' super cool and realistic.
Demo by @OzzyGT (Must try):
OzzyGT/diffusers-fast-inpaint
Its' super cool and realistic.
Demo by @OzzyGT (Must try):
OzzyGT/diffusers-fast-inpaint
Post
3639
Mistral Nemo is better than many models in 1st grader level reasoning.
Post
3973
I am experimenting with Flux and trying to push it to its limits without training (as I am GPU-poor 😅).
I found some flaws in the pipelines, which I resolved, and now I am able to generate an approx similar quality image as Flux Schnell 4 steps in just 1 step.
Demo Link:
KingNish/Realtime-FLUX
I found some flaws in the pipelines, which I resolved, and now I am able to generate an approx similar quality image as Flux Schnell 4 steps in just 1 step.
Demo Link:
KingNish/Realtime-FLUX
- 1 reply
Post
1968
I am excited to announce a major speed updated in Voicee, a superfast voice assistant.
It has now achieved latency <250 ms.
While its average latency is about 500ms.
KingNish/Voicee
This become Possible due to newly launched @sambanovasystems cloud.
You can also use your own API Key to get fastest speed.
You can get on from here: https://cloud.sambanova.ai/apis
For optimal performance use Google Chrome.
Please try Voicee and share your valuable feedback to help me further improve its performance and usability.
Thank you!
It has now achieved latency <250 ms.
While its average latency is about 500ms.
KingNish/Voicee
This become Possible due to newly launched @sambanovasystems cloud.
You can also use your own API Key to get fastest speed.
You can get on from here: https://cloud.sambanova.ai/apis
For optimal performance use Google Chrome.
Please try Voicee and share your valuable feedback to help me further improve its performance and usability.
Thank you!
Post
3632
Introducing Voicee, A superfast voice fast assistant.
KingNish/Voicee
It achieved latency <500 ms.
While its average latency is 700ms.
It works best in Google Chrome.
Please try and give your feedbacks.
Thank you. 🤗
KingNish/Voicee
It achieved latency <500 ms.
While its average latency is 700ms.
It works best in Google Chrome.
Please try and give your feedbacks.
Thank you. 🤗
- 3 replies
Post
5928
Introducing OpenCHAT mini: a lightweight, fast, and unlimited version of OpenGPT 4o.
KingNish/OpenCHAT-mini2
It has unlimited web search, vision and image generation.
Please take a look and share your review. Thank you! 🤗
KingNish/OpenCHAT-mini2
It has unlimited web search, vision and image generation.
Please take a look and share your review. Thank you! 🤗
-
- 7 replies
Post
15173
OpenGPT 4o now features WEB SEARCH
This feature enhances the capabilities of OpenGPT 4o, allowing it to fetch and integrate the latest information from the web directly into its responses.
Try Now: KingNish/OpenGPT-4o
With WEB SEARCH, OpenGPT 4o becomes an even more versatile and dynamic AI, ready to assist with up-to-date data retrieval and analysis.
This feature enhances the capabilities of OpenGPT 4o, allowing it to fetch and integrate the latest information from the web directly into its responses.
Try Now: KingNish/OpenGPT-4o
With WEB SEARCH, OpenGPT 4o becomes an even more versatile and dynamic AI, ready to assist with up-to-date data retrieval and analysis.
-
- 30 replies
Post
6504
I am pleased to announce 2 amazing AI demos:
1. Chat with Google Agent - This includes three AI models that allow you to converse with an AI, which provides answers by searching Google.
Demo Link: poscye/google-go
2. HelpingAI 9B - A model that surpassed all top AIs with the highest EQ benchmark score of 89.23. It specializes in understanding human emotions and responding in human style.
Demo Link: https://huggingface.co/spaces/Abhaykoul/HelpingAI-9B
Model Link: https://huggingface.co/OEvortex/HelpingAI-9B
Blog Link: https://huggingface.co/blog/KingNish/helpingai-9b
1. Chat with Google Agent - This includes three AI models that allow you to converse with an AI, which provides answers by searching Google.
Demo Link: poscye/google-go
2. HelpingAI 9B - A model that surpassed all top AIs with the highest EQ benchmark score of 89.23. It specializes in understanding human emotions and responding in human style.
Demo Link: https://huggingface.co/spaces/Abhaykoul/HelpingAI-9B
Model Link: https://huggingface.co/OEvortex/HelpingAI-9B
Blog Link: https://huggingface.co/blog/KingNish/helpingai-9b
-
- 2 replies
Post
3792
ChatGPT made Custom GPTs Free for Everyone.
Yes, you can use them but...
with limitations like
You can't use DallE 😥,
You can't make Custom GPTs
And chat limit also😥.
But...
We already have an open-source alternative like Hugging Chat, where you can create your custom assistant, generate, edit images, without any chat limit.
Try both of them from here:
https://chatgpt.com/gpts
https://huggingface.co/chat
and don't forget to Give your review here 👇:
Yes, you can use them but...
with limitations like
You can't use DallE 😥,
You can't make Custom GPTs
And chat limit also😥.
But...
We already have an open-source alternative like Hugging Chat, where you can create your custom assistant, generate, edit images, without any chat limit.
Try both of them from here:
https://chatgpt.com/gpts
https://huggingface.co/chat
and don't forget to Give your review here 👇:
- 4 replies
Post
2715
Introducing Image Generator Pro
KingNish/Image-Gen-Pro
It is Expert in Text to Image generation, Sequential Image generation or Image Editing.
Examples:
KingNish/Image-Gen-Pro
It is Expert in Text to Image generation, Sequential Image generation or Image Editing.
Examples:
-
-
- 5 replies
Post
2961
OpenGPT 4o NEW UPDATES:
1. Dedicated Image and Video Engine
2. Model Choices for Voice Chat
3. Better and Faster Voice Chat
4. Various Bug fixes
Test and give feedback of New features:
KingNish/OpenGPT-4o
Future Updates:
1. Web Search (Suggested by @GPT007 and @Saionton )
2. Live Chat with Voice Chat
3. Model Choices (Suggested by @NotAiLOL )
4. Multilingual Chats.
Suggest more features that should be added. 🤗
Thanks!
1. Dedicated Image and Video Engine
2. Model Choices for Voice Chat
3. Better and Faster Voice Chat
4. Various Bug fixes
Test and give feedback of New features:
KingNish/OpenGPT-4o
Future Updates:
1. Web Search (Suggested by @GPT007 and @Saionton )
2. Live Chat with Voice Chat
3. Model Choices (Suggested by @NotAiLOL )
4. Multilingual Chats.
Suggest more features that should be added. 🤗
Thanks!
-
-
- 6 replies
Post
4667
Microsoft Just Launched 3 Powerful Models
1. Phi 3 Medium (4k and 128k): A 14b Instruct tuned models that outperformed big models like Command R+ (104b), GPT 3.5 Pro, Gemini Pro, and is highly competitive with top models such as Mixtral 8x22b, Llama3 70B, and GPT 4.
microsoft/Phi-3-medium-4k-instruct
DEMO: https://huggingface.co/spaces/Walmart-the-bag/Phi-3-Medium
2. Phi 3 Mini Vision 128k: A 4.5 billion-parameter, instruction-tuned vision model that has outperformed models such as Llava3 and Claude 3, and is providing stiff competition to Gemini 1Pro Vision.
microsoft/Phi-3-vision-128k-instruct
3. Phi3 Small (8k and 128k): Better than Llama3 8b, Mixtral 8x7b and GPT 3.5 turbo.
microsoft/Phi-3-small-128k-instruct
1. Phi 3 Medium (4k and 128k): A 14b Instruct tuned models that outperformed big models like Command R+ (104b), GPT 3.5 Pro, Gemini Pro, and is highly competitive with top models such as Mixtral 8x22b, Llama3 70B, and GPT 4.
microsoft/Phi-3-medium-4k-instruct
DEMO: https://huggingface.co/spaces/Walmart-the-bag/Phi-3-Medium
2. Phi 3 Mini Vision 128k: A 4.5 billion-parameter, instruction-tuned vision model that has outperformed models such as Llava3 and Claude 3, and is providing stiff competition to Gemini 1Pro Vision.
microsoft/Phi-3-vision-128k-instruct
3. Phi3 Small (8k and 128k): Better than Llama3 8b, Mixtral 8x7b and GPT 3.5 turbo.
microsoft/Phi-3-small-128k-instruct
-
- 6 replies
Post
5081
Decoding GPT-4'o': Its Mechanisms and Creating Similar AI.
𝗥𝗲𝗮𝗱 𝗙𝘂𝗹𝗹 𝐀𝐫𝐭𝐢𝐜𝐥𝐞: https://huggingface.co/blog/KingNish/decoding-gpt-4o
𝐒𝐮𝐦𝐦𝐚𝐫𝐲 𝐨𝐟 𝐀𝐫𝐭𝐢𝐜𝐥𝐞- 📝
# 𝐌𝐞𝐜𝐡𝐚𝐧𝐢𝐜𝐬 𝐨𝐟 𝐆𝐏𝐓-𝟒’𝐨’: GPT-4’o’ operates through three main components 🛠️
𝟏. 𝐒𝐮𝐩𝐞𝐫𝐂𝐡𝐚𝐭: Integrates image generation, QnA (image, document and video) for diverse interactions.
𝟐. 𝐕𝐨𝐢𝐜𝐞 𝐂𝐡𝐚𝐭: Merges TTS and STT for real-time, human-like audio responses, focusing on human interaction.
𝟑. 𝐕𝐢𝐝𝐞𝐨 𝐂𝐡𝐚𝐭: Utilizes Zero Shot Image Classification to enhance user interaction with visual information.
# 𝐌𝐞𝐭𝐡𝐨𝐝𝐬 𝐭𝐨 𝐂𝐫𝐞𝐚𝐭𝐞 𝐒𝐢𝐦𝐢𝐥𝐚𝐫 𝐀𝐈 🧠
𝟏. 𝐌𝐮𝐥𝐭𝐢𝐌𝐨𝐝𝐚𝐥𝐢𝐟𝐢𝐜𝐚𝐭𝐢𝐨𝐧: Combines multiple models for a powerful, multifunctional AI.
𝟐. 𝐃𝐮𝐜𝐭 𝐓𝐚𝐩𝐞 𝐌𝐞𝐭𝐡𝐨𝐝: Uses different models or APIs for specific tasks without additional training.
The article provides an in-depth exploration of GPT-4’o’, its functionalities, and methods to create similar AI models. It emphasizes the model’s language support and its innovative approach to human-AI interaction. 💡🌐
(𝙉𝙊𝙏𝙀: 𝙎𝙪𝙢𝙢𝙖𝙧𝙮 𝙞𝙨 𝘼𝙄 𝙜𝙚𝙣𝙚𝙧𝙖𝙩𝙚𝙙) ✅
𝗥𝗲𝗮𝗱 𝗙𝘂𝗹𝗹 𝐀𝐫𝐭𝐢𝐜𝐥𝐞: https://huggingface.co/blog/KingNish/decoding-gpt-4o
𝐒𝐮𝐦𝐦𝐚𝐫𝐲 𝐨𝐟 𝐀𝐫𝐭𝐢𝐜𝐥𝐞- 📝
# 𝐌𝐞𝐜𝐡𝐚𝐧𝐢𝐜𝐬 𝐨𝐟 𝐆𝐏𝐓-𝟒’𝐨’: GPT-4’o’ operates through three main components 🛠️
𝟏. 𝐒𝐮𝐩𝐞𝐫𝐂𝐡𝐚𝐭: Integrates image generation, QnA (image, document and video) for diverse interactions.
𝟐. 𝐕𝐨𝐢𝐜𝐞 𝐂𝐡𝐚𝐭: Merges TTS and STT for real-time, human-like audio responses, focusing on human interaction.
𝟑. 𝐕𝐢𝐝𝐞𝐨 𝐂𝐡𝐚𝐭: Utilizes Zero Shot Image Classification to enhance user interaction with visual information.
# 𝐌𝐞𝐭𝐡𝐨𝐝𝐬 𝐭𝐨 𝐂𝐫𝐞𝐚𝐭𝐞 𝐒𝐢𝐦𝐢𝐥𝐚𝐫 𝐀𝐈 🧠
𝟏. 𝐌𝐮𝐥𝐭𝐢𝐌𝐨𝐝𝐚𝐥𝐢𝐟𝐢𝐜𝐚𝐭𝐢𝐨𝐧: Combines multiple models for a powerful, multifunctional AI.
𝟐. 𝐃𝐮𝐜𝐭 𝐓𝐚𝐩𝐞 𝐌𝐞𝐭𝐡𝐨𝐝: Uses different models or APIs for specific tasks without additional training.
The article provides an in-depth exploration of GPT-4’o’, its functionalities, and methods to create similar AI models. It emphasizes the model’s language support and its innovative approach to human-AI interaction. 💡🌐
(𝙉𝙊𝙏𝙀: 𝙎𝙪𝙢𝙢𝙖𝙧𝙮 𝙞𝙨 𝘼𝙄 𝙜𝙚𝙣𝙚𝙧𝙖𝙩𝙚𝙙) ✅
- 2 replies