language:

• en
  license: other
  base_model:
• huihui-ai/Huihui-Qwen3.5-9B-abliterated
  tags:
• gptq
• 4bit
• quantized
• qwen
• text-generation
• gptq-pro
  pipeline_tag: text-generation
  library_name: transformers

Huihui-Qwen3.5-9B-abliterated GPTQ-Pro 4bit (g64)

Overview

Huihui-Qwen3.5-9B-abliterated-GPTQ-Pro-4bit-g64 is a GPTQ-quantized checkpoint intended for efficient GPU inference, published by groxaxo.
It is intended for open-source evaluation, reproducible experimentation, and compatible local or
hosted inference workflows. The wording below is deliberately limited to what can be verified
from this repository's metadata and artifacts.

The repository name identifies a behavior-modified or reduced-filtering lineage. That label describes the source or conversion history; it is not a guarantee of unrestricted behavior in every prompt or runtime. Test outputs carefully before sharing or deploying them.

At a glance

What is included

• *.safetensors (2 files)
• config.json
• generation_config.json
• tokenizer.json
• tokenizer_config.json
• chat_template.jinja
• quantize_config.json
• Additional configuration, tokenizer, processor, or shard files (10 visible artifacts total)

Quick start

vLLM (documented configuration)

vllm serve groxaxo/Huihui-Qwen3.5-9B-abliterated-GPTQ-Pro-4bit-g64 \
  --quantization gptq_marlin \
  --dtype float16 \
  --trust-remote-code

This command is taken from the repository documentation. Adjust tensor parallelism, context
length, and cache settings to match your hardware and vLLM version.

Compatibility and responsible use

• Use a runtime that explicitly supports this format, architecture, and modality.
• Keep configuration, tokenizer, processor, projection, and weight files from the same revision together.
• Review the source model card and license before redistribution or deployment.
• Hardware needs depend on parameter count, context length, cache precision, quantization, and concurrency.
• Report reproducible issues with the runtime version, hardware, launch command, and a minimal example.

Quantization or conversion changes numerical behavior, memory use, and throughput relative to the source checkpoint; validate quality on your own workload.

Generated outputs may be inaccurate or unsuitable for a given use case. Users are responsible for
testing behavior, applying appropriate safeguards, and complying with applicable licenses and laws.

This is a GPTQ-Pro 4-bit quantization of huihui-ai/Huihui-Qwen3.5-9B-abliterated.

It was quantized with group size 64 and evaluated against the original model on Wikitext-2 using a strided perplexity setup, plus KL and token-agreement checks.

Highlights

• Base model: huihui-ai/Huihui-Qwen3.5-9B-abliterated
• Quantization: GPTQ-Pro, 4-bit, group size 64
• Calibration samples: 128
• Quantization time: about 11.1 minutes
• Quantized strided perplexity: 9.6579
• Original strided perplexity: 9.5234
• Perplexity degradation: 1.41%
• Average KL divergence vs original: 0.03423
• Top-1 agreement vs original: 91.96%
• Top-5 agreement vs original: 99.98%

Quality Notes

This quantized build stays very close to the source model in language modeling quality.

• Perplexity regression is small.
• KL divergence is low.
• Top-5 next-token agreement is effectively perfect.
• In practice, this should preserve most of the original model's behavior while reducing memory use substantially.

Files

• model-00001-of-00002.safetensors
• model-00002-of-00002.safetensors
• quantize_config.json
• tokenizer and config files

Load With Transformers / GPTQModel

from gptqmodel import GPTQModel

model = GPTQModel.load(
    "groxaxo/Huihui-Qwen3.5-9B-abliterated-GPTQ-Pro-4bit-g64",
    device_map="auto",
    trust_remote_code=True,
)

Evaluation Summary

Measured locally:

• Quantized strided PPL: 9.6579304371
• Original strided PPL: 9.5233634665
• Quantized chunked PPL: 11.6689118281
• Original chunked PPL: 11.5080707440
• KL divergence: 0.0342324856
• Logit cosine similarity: 0.9935612157

Prompting

Use the same prompting and chat template behavior as the base model.

Disclaimer

This repo contains only the quantized checkpoint. Please review the base model card for intended use, limitations, and licensing details.