Extend TORCH_CUDA_ARCH_LIST to support additional CUDA architectures, including sm_80, sm_86, sm_87, sm_89, and sm_90

[maxmet91]

Current Build Support:

• Pascal (GeForce 10 series, P100)
• Turing (GeForce 20 series, V100)
• Supported CUDA capabilities: sm_60, sm_70, sm_75, compute_70, compute_75.

Issue: On newer CUDA architectures, starting from Ampere (GeForce 30 series, A100), there is a warning message when PyTorch with CUDA (GPU) is being used:

/opt/conda/lib/python3.10/site-packages/torch/cuda/__init__.py:215: UserWarning: 
NVIDIA GeForce RTX 3070 Laptop GPU with CUDA capability sm_86 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_60 sm_70 sm_75 compute_70 compute_75.

For some tasks, this leads to FATAL ERRORs and makes GPU utilization impossible:

FATAL: kernel `fmha_cutlassF_f32_aligned_64x64_rf_sm80` is for sm80-sm100, but was built for sm75

Objective of this Pull Request: This Pull Request aims to address the compatibility issues and extend support for all architectures up to Ampere (GeForce 40 series, H100).

[google-cla[bot]]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[calderjo]

looks like it build without failure, but not 100% sure this is for.

please explain the use case and update the description

[maxmet91]

looks like it build without failure, but not 100% sure this is for.

please explain the use case and update the description

I updated the description. Please check. This update is intended for using the gpu docker image on hardware other than Kaggle's.

[calderjo]

we don't offer support for other software