build(torch)!: Drop arches 6.0, 6.1, 6.2, & 7.2 as compilation targets

[Eta0]

Slimmer PyTorch Builds

Compilation target architectures 6.0, 6.1, 6.2, & 7.2 do not correspond to any GPUs currently offered by CoreWeave, and bloat the resulting builds, so they are removed as CI defaults. They can still be compiled by manually invoking a Dockerfile build.

The list of GPU architectures currently offered by CoreWeave are:

GPU	Generation	Architecture Code
H100_NVLINK_80GB	Hopper	sm_90
H100_PCIE	Hopper	sm_90
A100_NVLINK_80GB	Ampere	sm_80
A100_NVLINK	Ampere	sm_80
A100_PCIE_40GB	Ampere	sm_80
A100_PCIE_80GB	Ampere	sm_80
A40	Ampere	sm_86
RTX_A6000	Ampere	sm_86
RTX_A5000	Ampere	sm_86
RTX_A4000	Ampere	sm_86
Tesla_V100_NVLINK	Volta	sm_70
Quadro_RTX_5000	Turing	sm_75
Quadro_RTX_4000	Turing	sm_75

[github-actions[bot]]

@Eta0 Build complete, success: https://github.com/coreweave/ml-containers/actions/runs/5016929748 Image: ghcr.io/coreweave/ml-containers/torch:es-slim-torch-6743aea-base-cuda11.8.0-torch2.0.0-vision0.15.1-audio2.0.1