Skip to content
Back to pull request #1670

Support power of 2 scaling factors in float8 training and use e4m3 everywhere #7884

Sign in to view logs

Support power of 2 scaling factors in float8 training and use e4m3 everywhere

Support power of 2 scaling factors in float8 training and use e4m3 everywhere #7884

test-nightly (CUDA Nightly, linux.g5.12xlarge.nvidia.gpu, --pre torch==2.7.0.dev20250122 --index-...  /  linux-job

test-nightly (CUDA Nightly, linux.g5.12xlarge.nvidia.gpu, --pre torch==2.7.0.dev20250122 --index-... / linux-job
succeeded Feb 6, 2025 in 47m 35s