About this repo: AWQ W4A16 quantization

This is a 4-bit AWQ (W4A16, asymmetric, group size 128) quantization of huihui-ai/Huihui-ThinkingCap-Qwen3.6-27B-abliterated, produced with LLM-Compressor (compressed-tensors pack-quantized format).

Requirements: vLLM >= 0.25 (the config uses the composite Qwen3_5ForConditionalGeneration architecture; quantized GDN/linear-attention projections rely on Marlin thread-tile padding available in 0.25+).

Repo layout:

model.safetensors — quantized language model weights (all Linear layers, incl. linear-attention)

vision.safetensors — vision tower in bf16 (unquantized; image/video input works out of the box)

model-base-aux.safetensors — MTP weights in bf16 for speculative decoding

Example (vLLM):

bash
1vllm serve DoktorMincs/Huihui-ThinkingCap-Qwen3.6-27B-abliterated-AWQ-W4A16 \ 2 --tensor-parallel-size 4 --max-model-len 163840 \ 3 --tool-call-parser qwen3_coder --reasoning-parser qwen3 --enable-auto-tool-choice \ 4 --enable-prefix-caching --max-num-batched-tokens 8192 \ 5 --speculative-config '{"method": "mtp", "num_speculative_tokens": 3}'

Tested on 4x RTX 3060 12GB (TP=4): ~71 tok/s single-stream decode with MTP (draft acceptance ~63%), 4K image input verified. lm_head, vision tower and MTP weights stay in bf16.

Original model card below (from the source repo):

huihui-ai/Huihui-ThinkingCap-Qwen3.6-27B-abliterated

This is an uncensored version of bottlecapai/ThinkingCap-Qwen3.6-27B created with abliteration (see remove-refusals-with-transformers to know more about it). This is a crude, proof-of-concept implementation to remove refusals from an LLM model without using TransformerLens.

Usage Warnings

Risk of Sensitive or Controversial Outputs: This model’s safety filtering has been significantly reduced, potentially generating sensitive, controversial, or inappropriate content. Users should exercise caution and rigorously review generated outputs.
Not Suitable for All Audiences: Due to limited content filtering, the model’s outputs may be inappropriate for public settings, underage users, or applications requiring high security.
Legal and Ethical Responsibilities: Users must ensure their usage complies with local laws and ethical standards. Generated content may carry legal or ethical risks, and users are solely responsible for any consequences.
Research and Experimental Use: It is recommended to use this model for research, testing, or controlled environments, avoiding direct use in production or public-facing commercial applications.
Monitoring and Review Recommendations: Users are strongly advised to monitor model outputs in real-time and conduct manual reviews when necessary to prevent the dissemination of inappropriate content.
No Default Safety Guarantees: Unlike standard models, this model has not undergone rigorous safety optimization. huihui.ai bears no responsibility for any consequences arising from its use.

Donation

Your donation helps us continue our further development and improvement, a cup of coffee can do it.

bitcoin:

bc1qqnkhuchxw0zqjh2ku3lu4hq45hc6gy84uk70ge

Support our work on Ko-fi!