[NVIDIA] Generate aarch64 binaries for blackwell family #161378
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/161378
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit b7a9e6a with merge base 74280d0 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@pytorchbot label "topic: not user facing" |
| GPU_ARCH_VERSION=${GPU_ARCH_VERSION:-} | ||
|
|
||
| if [[ "$GPU_ARCH_VERSION" == *"12.9"* ]]; then | ||
| export TORCH_CUDA_ARCH_LIST="8.0;9.0;10.0;12.0" |
There was a problem hiding this comment.
10.3 is supported in 12.9 fwiw :) it seems to work
There was a problem hiding this comment.
Yes, also spark. The only really new is Thor eith OpenRM driver in cuda 13
There was a problem hiding this comment.
maybe dumb question, why only add for 13 not 12.9? downstream code will still need to bump for cu13 too, no?
|
This PR cannot land as it's non needed and I would recommend trying to understand our binary compatibility support in the CUDA toolkit before submitting these changes @johnnynunez. |
Got it! |
4 participants
Include Thor, Spark and GB300. When Orin be SBSA, we can include it
Fixes #161377
@Skylion007 @tinglvv @ptrblck @nWEIdia