量化qwen3-8b

[Jim2016713]

使用配置量化awq_w8a8.yml量化qwen3-8b时报了这样的错
[rank0]: Traceback (most recent call last):
[rank0]: File "/code/llm_quantization/LightCompress//llmc/main.py", line 261, in
[rank0]: main(config)
[rank0]: File "/code/llm_quantization/LightCompress//llmc/main.py", line 69, in main
[rank0]: blockwise_opt.run_block_loop()
[rank0]: File "/code/llm_quantization/LightCompress/llmc/compression/blockwise_optimization.py", line 38, in run_block_loop
[rank0]: self.block_opt(self.blocks[self.block_idx])
[rank0]: File "/code/llm_quantization/LightCompress/llmc/compression/quantization/base_blockwise_quantization.py", line 416, in block_opt
[rank0]: self.run(block, input_feat, handles)
[rank0]: File /code/llm_quantization/LightCompress/llmc/compression/quantization/base_blockwise_quantization.py", line 447, in run
[rank0]: self.block_transform(block, input_feat, self.input['kwargs'])
[rank0]: File "/opt/miniconda/envs/light-compress/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 124, in decorate_context
[rank0]: return func(*args, **kwargs)
[rank0]: ^^^^^^^^^^^^^^^^^^^^^
[rank0]: File "/code/llm_quantization/LightCompress/llmc/compression/quantization/awq.py", line 283, in block_transform
[rank0]: super().block_transform(block, input_feat, block_kwargs)
[rank0]: File "/code/llm_quantization/LightCompress/llmc/compression/quantization/base_blockwise_quantization.py", line 488, in block_transform
[rank0]: self.subset_transform(
[rank0]: File "/opt/miniconda/envs/light-compress/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 124, in decorate_context
[rank0]: return func(*args, **kwargs)
[rank0]: ^^^^^^^^^^^^^^^^^^^^^
[rank0]: File "/code/llm_quantization/LightCompress/llmc/compression/quantization/awq.py", line 355, in subset_transform
[rank0]: scale = self.search_scale_subset(
[rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]: File "/opt/miniconda/envs/light-compress/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 124, in decorate_context
[rank0]: return func(*args, **kwargs)
[rank0]: ^^^^^^^^^^^^^^^^^^^^^
[rank0]: File "/code/llm_quantization/LightCompress/llmc/compression/quantization/awq.py", line 272, in search_scale_subset
[rank0]: best_scales = torch.zeros_like(best_scales, device='cuda')
[rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]: TypeError: zeros_like(): argument 'input' (position 1) must be Tensor, not NoneType