Hello everyone. My previous post in this thread on reddit recieved a lot of upvotes and warm and great feedback. Thank you very much guys. So I decided to improve and refine my workflow even further via merging more Qwen 3.5 9B models this time.
Introducing fully uncensored Omnicoder distilled by Claude Opus with zero refusals:
https://huggingface.co/LuffyTheFox/Omnicoder-Claude-4.6-Opus-Uncensored-GGUF
Q4_K_M quants and Q8_0 quants now availble. I tested only Q8_O quant. It's pretty solid.
It has been made by Add Difference merge via this script: https://pastebin.com/xEP68vss
I preserved GGUF header and metadata structure for compability.
Frankly saying I was surpised how ... stupid Claude Opus 4.6 is. It broke this simple Python script almost 10 times when i asked him to add huggingface upload feature and chat template change feature in GGUF file.
So my merge has been made via following models:
1) Latest update for Jackrong model trained on distilled dataset from Claude Opus: https://huggingface.co/Jackrong/Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distilled-v2-GGUF
2) HauhauCS uncensored Qwen 3.5 9B model https://huggingface.co/HauhauCS/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive
3) Omnicoder made by Tesslate: https://huggingface.co/Tesslate/OmniCoder-9B-GGUF
4) And i used Bartowski quant as base: https://huggingface.co/bartowski/Qwen_Qwen3.5-9B-GGUF
I think it's best thing what we have now in terms of UGI (Uncensored General Intelligence) for small 9B model based on Qwen 3.5 9B architecture.
Feel free to test it in Open Claw and share your results.
[link] [comments]
