models icon indicating copy to clipboard operation
models copied to clipboard
verified publisher icon intel

RNNT training on CPU

Open Peach-He opened this issue 4 years ago • 1 comments

Appreciate for the job on supporting RNN-T training on CPU (models/language_modeling/pytorch/rnnt/training/cpu), just quick evaluated the training code and found that WER would keep in 1.00 after even training 10+ epoches. And I found this issue related on loss function used in training https://github.com/HawkAaron/warp-transducer/issues/93 The grad in cpu is incorrect, is this a know issue? Or have we ever gotten the final WER of 0.058 rather than 1.0?

Peach-He avatar Jan 20 '22 09:01 Peach-He

@Peach-He: The RNNT CPU training scripts have been updated recently. Can you try again to see if it resolves your issues? You can refer to latest optimizations here: https://www.intel.com/content/www/us/en/developer/articles/containers/cpu-reference-model-containers.html

sramakintel avatar Mar 25 '24 16:03 sramakintel