Is it possible to remove fixed torch version?

[qinxuye]

Now the version of pytorch is fixed to 2.3.0, but many libraries like vllm and sglang has upgraded to torch 2.4.0, there is many conflicts when we want to include them and AutoAWQ in a docker, is it possible not to fix the torch version?

[casper-hansen]

I want to make it more flexible, but PyPi only allow for one wheel. So I cannot upload multiple versions, but one fix could be to implement a flag in AutoAWQ to avoid installing AutoAWQ_kernels since you do not need them for quantization.

[casper-hansen]

@qinxuye would it help with a flag like this? https://github.com/casper-hansen/AutoAWQ/pull/582

[stenreijers]

This can be closed too due to the merge of #582

[frei-x]

+1

[liangjs]

This can be closed too due to the merge of #582

Please release AutoAWQ with this fix as a new version, because the current version asks for a specific version of pytorch, which can break dependencies of other packages.