search

QwenLM / Qwen2-VL

Qwen2-VL is the multimodal large language model series developed by Qwen team, Alibaba Cloud.
Apache License 2.0
2.02k stars 116 forks source link

AttributeError: 'Qwen2VLForConditionalGeneration' object has no attribute 'quantize'. Did you mean: 'dequantize'? #208

Open CachCheng opened 5 days ago

CachCheng commented 5 days ago

[INFO:swift] Qwen2VLForConditionalGeneration: 8291.3756M Params (0.0000M Trainable [0.0000%]), 234.8828M Buffers. [INFO:swift] system: You are a helpful assistant. [INFO:swift] Quantization dataset: ['ms-bench'] [INFO:swift] Start quantizing the model... Traceback (most recent call last): File "/meta/cash/llm/swift/swift/cli/export.py", line 5, in export_main() File "/meta/cash/llm/swift/swift/utils/run_utils.py", line 32, in x_main result = llm_x(args, **kwargs) File "/meta/cash/llm/swift/swift/llm/export.py", line 252, in llm_export awq_model_quantize(model, template.tokenizer, args.quant_batch_size) File "/meta/cash/llm/swift/swift/llm/export.py", line 138, in awq_model_quantize awq_model.quantize(tokenizer, quant_config=quant_config, n_parallel_calib_samples=batch_size) File "/home/ahs/anaconda3/envs/py310/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1729, in getattr raise AttributeError(f"'{type(self).name}' object has no attribute '{name}'") AttributeError: 'Qwen2VLForConditionalGeneration' object has no attribute 'quantize'. Did you mean: 'dequantize'?

kq-chen commented 1 day ago

Hi, please kindly report this to swift.