Feat: Add torch to requirements

[NanoCode012]

Error with pip install -e . in new venv due to missing torch.

[winglian]

torch gets installed in the base docker env, but it definitely should exists in requirements. But I think when pip install runs, it tries to delete the version that's already installed. thoughts?

[NanoCode012]

Hm, when I ran pip install -e . in my docker image. It does not install torch again.

Requirement already satisfied: torch>=1.6.0 in /root/miniconda3/envs/py3.9/lib/python3.9/site-packages (from accelerate) (2.0.0)

[winglian]

thanks, what about different version of torch? for example we build images with 1.13.1 and 2.0.0
https://github.com/OpenAccess-AI-Collective/axolotl/blob/main/.github/workflows/base.yml#L17-L22

[NanoCode012]

I will build the Dockerfile following your dev image for 1.13.1 from docker hub and let you know.

[NanoCode012]

You're right. It does try to install torch 2.0

[NanoCode012]

Shall we update the Readme to require Pytorch installed and add a commented to QuickStart

# pip3 install torch

and requirements?

# torch

[winglian]

Shall we update the Readme to require Pytorch installed and add a commented to QuickStart

# pip3 install torch

and requirements?

# torch

yeah, I think adding it to the docs is sufficient, something like if torch isn't installed yet...

[winglian]

what about torch>=1.13.1 ?

[NanoCode012]

what about torch>=1.13.1 ?

=> => # Collecting torch>=1.13.1 (from axolotl==0.1)                                                                                                                                                                 
 => => #   Downloading torch-2.0.1-cp39-cp39-manylinux1_x86_64.whl (619.9 MB)