Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-GPTQ-int4

This is a GPTQ INT4 quantized version of Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled.

Please refer to the original model card for details on the model architecture, training data, and capabilities.

Note: While the original fine-tuning focused on text-only reasoning tasks, this model inherits multimodal capabilities from the base Qwen3.5-27B. The vision encoder is preserved and functional for image understanding tasks.

Quantization Details

• Method: GPTQ (4-bit INT4, W4A16)
• Group Size: 128
• Calibration: 1024 samples from C4 dataset
• Vision Encoder: Preserved (not quantized)
• MTP Module: Preserved (not quantized)

Usage with vLLM

Text-only

from vllm import LLM, SamplingParams

llm = LLM(
    model="codgician/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-GPTQ-int4",
    trust_remote_code=True,
    max_model_len=4096,
    gpu_memory_utilization=0.9,
)

sampling_params = SamplingParams(temperature=0.7, max_tokens=2048)
prompt = "Explain the difference between TCP and UDP protocols."
outputs = llm.generate([prompt], sampling_params)
print(outputs[0].outputs[0].text)

With Image (Multimodal)

from vllm import LLM, SamplingParams

llm = LLM(
    model="codgician/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-GPTQ-int4",
    trust_remote_code=True,
    max_model_len=4096,
    gpu_memory_utilization=0.9,
)

sampling_params = SamplingParams(temperature=0.7, max_tokens=256)
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image_url", "image_url": {"url": "https://example.com/image.jpg"}},
            {"type": "text", "text": "What is in this image?"}
        ]
    }
]
outputs = llm.chat(messages, sampling_params)
print(outputs[0].outputs[0].text)

Hardware Requirements

Precision	VRAM (Approx.)
INT4 GPTQ	~18 GB

Acknowledgements

• Original model by Jackrong
• Base model: Qwen/Qwen3.5-27B
• Quantization performed using GPTQModel

License

Apache 2.0 (inherited from original model)