Faster RNN-T Label-Looping Decoding with WIND

[artbataev]

[!IMPORTANT]
The Update branch button must only be pressed in very rare occassions. An outdated branch is never blocking the merge of a PR. Please reach out to the automation team before pressing that button.

What does this PR do ?

Implementation of WIND algorithm with label-looping for faster RNN-T decoding. For algorithm details see "WIND: Accelerated RNN-T Decoding with Windowed Inference for Non-blank Detection" https://arxiv.org/abs/2505.13765

Collection: [ASR]

Changelog

• Add specific line by line info of high level changes in this PR.

Usage

python examples/asr/speech_to_text_eval.py \
 pretrained_name=stt_en_fastconformer_transducer_large \
 dataset_manifest=<manifest> \
 batch_size=32 \
 output_filename=results.jsonl \
 amp=false \
 compute_dtype=bfloat16 \
 matmul_precision=high \
 presort_manifest=true \
 calculate_rtfx=true \
 warmup_steps=1 \
 rnnt_decoding.strategy=greedy_batch \
 rnnt_decoding.greedy.window_size=8

GitHub Actions CI

The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.

The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR. To re-run CI remove and add the label again. To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".

Before your PR is "Ready for review"

Pre checks:

• [x] Make sure you read and followed Contributor guidelines
• [x] Did you write any new necessary tests?
• [ ] Did you add or update any necessary documentation?
• [ ] Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • [ ] Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• [x] New Feature
• [ ] Bugfix
• [ ] Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed. Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

• Related to # (issue)

[MahmoudAshraf97]

Great work, this brings TDT speeds to RNNT models without the need to retrain, was there any work comparing triton kernels to cuda graphs for batched decoding?

[github-actions[bot]]

This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.

[github-actions[bot]]

This PR was closed because it has been inactive for 7 days since being marked as stale.

[hlevring]

Was wondering about the status of this PR. Don't see any reviewers with comments yet, so I guess it could still take some time before it get merged.