this was helpful - but IIUC torch==1.13.1 also works

Hey thanks for the PR. So far we only put the most concerning version control in the requirement.txt, which is transformers. For other packages, it may just work for several versions. I will merge this PR in the future if more issues show that versioning plays a big role in optimization.

Thanks.

The existing requirements.txt wasn't enough for me and I had to manually run pip install torch.

Yeah you need to install pytorch. There are several pytorch versions for different CUDA versions so I incline to leave it to users to decide based on their own machine.

When I ran pip install -e . it gave me the following error:

INFO: pip is looking at multiple versions of fschat to determine which version is compatible with other requirements. This could take a while.
ERROR: Cannot install llm-attacks and llm-attacks==0.0.1 because these package versions have conflicting dependencies.

The conflict is caused by:
    llm-attacks 0.0.1 depends on transformers==4.28.1
    fschat 0.2.23 depends on transformers>=4.31.0

To fix this you could try to:
1. loosen the range of package versions you've specified
2. remove package versions to allow pip attempt to solve the dependency conflict

ERROR: ResolutionImpossible: for help visit https://pip.pypa.io/en/latest/topics/dependency-resolution/#dealing-with-dependency-conflicts

The same thing happened when I attempted to install both llm-attacks and fastchat manually.

@tjaffee99 Solved. Thanks for pointing out the issue. We did not have this issue before as fschat was not in the requirements.txt. I pushed new versions in the requirements.txt