[Bug] RuntimeError: Boolean value of Tensor with more than one value is ambiguous

[ChenMnZ]

I strictly follow the instruction of environment installation. However, when I run the following command:

python -m lmquant.llm.run configs/llm.yaml configs/qoq/gchn.yaml --model-name llama2-7b --smooth-xw-alpha 0 --smooth-xw-beta 1

I still meet the following error:

  File "/cpfs01/user/chenmengzhao/efficientqat_repo/lmquant/lmquant/llm/run.py", line 276, in <module>
    run(config)
  File "/cpfs01/user/chenmengzhao/efficientqat_repo/lmquant/lmquant/llm/run.py", line 155, in run
    smooth_llm(model, config.quant, smooth_cache=torch.load(config.cache_path.smooth))
  File "/cpfs01/user/chenmengzhao/anaconda3/envs/lmquant/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/cpfs01/user/chenmengzhao/efficientqat_repo/lmquant/lmquant/llm/quant/smooth.py", line 189, in smooth_llm
    smooth_cache = smooth_cache or {}
                   ^^^^^^^^^^^^^^^^^^
RuntimeError: Boolean value of Tensor with more than one value is ambiguous

[xpq-tech]

Have you solved it? I also encountered this problem.

[ChenMnZ]

@xpq-tech I still not know how to solve this problem.

[JamesTheZ]

Had a similar problem and solved by downgrading the transformer version to 4.42.0.