GPTQ quantization issue while doing inference in vLLM after swift export

[Dineshkumar-Anandan-ZS0367]

ValueError: The input size is not aligned with the quantized weight shape. This can be caused by too large tensor parallel size.

Model name - Ovis2.5 9B
Method - gptq

[Jintao-Huang]

If it runs fine with transformers, please open an issue for vllm.