Skip to content

Make FloatToFloat16 conversion 75x faster using SVE2 instructions #8859

Make FloatToFloat16 conversion 75x faster using SVE2 instructions

Make FloatToFloat16 conversion 75x faster using SVE2 instructions #8859

Triggered via pull request January 29, 2025 02:26
Status Failure
Total duration 3h 2m 42s
Artifacts 5

build_wheels_linux_x86.yml

on: pull_request
generate-matrix  /  generate
7s
generate-matrix / generate
Matrix: pytorch/FBGEMM / build
Matrix: build / upload / upload
Fit to window
Zoom out
Zoom in

Annotations

3 errors
pytorch/FBGEMM / manywheel-py3_9-rocm6_3
The job running on runner i-0b77d277000a42f40 has exceeded the maximum execution time of 180 minutes.
pytorch/FBGEMM / manywheel-py3_9-rocm6_3
The operation was canceled.
pytorch/FBGEMM / upload / manywheel-py3_9-rocm6_3
Unable to download artifact(s): Artifact not found for name: pytorch_FBGEMM__3.9_rocm6.3_x86_64 Please ensure that your artifact is not expired and the artifact was uploaded using a compatible version of toolkit/upload-artifact. For more information, visit the GitHub Artifacts FAQ: https://github.com/actions/toolkit/blob/main/packages/artifact/docs/faq.md

Artifacts

Produced during runtime
Name Size
pytorch_FBGEMM__3.9_cpu_x86_64
3.75 MB
pytorch_FBGEMM__3.9_cu118_x86_64
342 MB
pytorch_FBGEMM__3.9_cu124_x86_64
452 MB
pytorch_FBGEMM__3.9_cu126_x86_64
449 MB
pytorch_FBGEMM__3.9_rocm6.2.4_x86_64
127 MB