TPS-free 2D bucket estimation and filtering#11738
Conversation
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
| from lhotse.testing.dummies import DummyManifest, dummy_cut | ||
| from nemo.collections.common.data.lhotse.sampling import FixedBucketBatchSizeConstraint2D | ||
| from lhotse.testing.dummies import dummy_cut | ||
| from lhotse.testing.random import deterministic_rng |
Check notice
Code scanning / CodeQL
Unused import Note test
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: pzelasko <pzelasko@users.noreply.github.com>
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
…com/nvidia/nemo into 2d-bucketing-and-tps-improvements-2
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
|
|
||
| if config.get("tarred_random_access", False): | ||
| logging.warning( | ||
| "Option 'tarred_random_access' is deprecated and replaced with 'skip_missing_manifest_entries'.", |
There was a problem hiding this comment.
may be also add version from which this would be removed
Signed-off-by: Piotr Zelasko <pzelasko@cs-oci-ord-login-01.cm.cluster>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
49a1536 to
a238da2
Compare
Signed-off-by: Piotr Żelasko <petezor@gmail.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
|
beep boop 🤖: 🚨 The following files must be fixed before merge! Your code was analyzed with PyLint. The following annotations have been identified: Mitigation guide:
By applying these rules, we reduce the occurance of this message in future. Thank you for improving NeMo's documentation! |
|
[🤖]: Hi @pzelasko 👋, We wanted to let you know that a CICD pipeline for this PR just finished successfully So it might be time to merge this PR or get some approvals I'm just a bot so I'll leave it you what to do next. //cc @pablo-garay @ko3n1g |
* TPS-free 2D bucket estimation and filtering Signed-off-by: Piotr Żelasko <petezor@gmail.com> * fixes Signed-off-by: Piotr Żelasko <petezor@gmail.com> * extra unit test Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Fixes in 2D bin estimation script and OOMptimizer Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> * Apply isort and black reformatting Signed-off-by: pzelasko <pzelasko@users.noreply.github.com> * Support list values for max_tps/max_tpt and bucketing Signed-off-by: Piotr Żelasko <petezor@gmail.com> * fix max_tps/max_tpt config parsing for list values Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> * Transition: tarred_random_access -> skip_missing_manifest_entries Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Improve warning messages Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Support single unified BPE tokenizer for Canary2 Signed-off-by: Piotr Zelasko <pzelasko@cs-oci-ord-login-01.cm.cluster> * bugfix +es to unified tokenizer support for Canary Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> * CanaryBPETokenizer Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Revert changes to fetching pad id in AED models Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> * Update docs Signed-off-by: Piotr Żelasko <petezor@gmail.com> * fix tests Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Remove max_tps and max_duration from OOMptimizer output Signed-off-by: Piotr Żelasko <petezor@gmail.com> --------- Signed-off-by: Piotr Żelasko <petezor@gmail.com> Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> Signed-off-by: pzelasko <pzelasko@users.noreply.github.com> Signed-off-by: Piotr Zelasko <pzelasko@cs-oci-ord-login-01.cm.cluster> Co-authored-by: pzelasko <pzelasko@users.noreply.github.com> Signed-off-by: Parth Mannan <pmannan@nvidia.com>
* TPS-free 2D bucket estimation and filtering Signed-off-by: Piotr Żelasko <petezor@gmail.com> * fixes Signed-off-by: Piotr Żelasko <petezor@gmail.com> * extra unit test Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Fixes in 2D bin estimation script and OOMptimizer Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> * Apply isort and black reformatting Signed-off-by: pzelasko <pzelasko@users.noreply.github.com> * Support list values for max_tps/max_tpt and bucketing Signed-off-by: Piotr Żelasko <petezor@gmail.com> * fix max_tps/max_tpt config parsing for list values Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> * Transition: tarred_random_access -> skip_missing_manifest_entries Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Improve warning messages Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Support single unified BPE tokenizer for Canary2 Signed-off-by: Piotr Zelasko <pzelasko@cs-oci-ord-login-01.cm.cluster> * bugfix +es to unified tokenizer support for Canary Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> * CanaryBPETokenizer Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Revert changes to fetching pad id in AED models Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> * Update docs Signed-off-by: Piotr Żelasko <petezor@gmail.com> * fix tests Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Remove max_tps and max_duration from OOMptimizer output Signed-off-by: Piotr Żelasko <petezor@gmail.com> --------- Signed-off-by: Piotr Żelasko <petezor@gmail.com> Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> Signed-off-by: pzelasko <pzelasko@users.noreply.github.com> Signed-off-by: Piotr Zelasko <pzelasko@cs-oci-ord-login-01.cm.cluster> Co-authored-by: pzelasko <pzelasko@users.noreply.github.com> Signed-off-by: Abhinav Garg <abhgarg@nvidia.com>
* TPS-free 2D bucket estimation and filtering Signed-off-by: Piotr Żelasko <petezor@gmail.com> * fixes Signed-off-by: Piotr Żelasko <petezor@gmail.com> * extra unit test Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Fixes in 2D bin estimation script and OOMptimizer Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> * Apply isort and black reformatting Signed-off-by: pzelasko <pzelasko@users.noreply.github.com> * Support list values for max_tps/max_tpt and bucketing Signed-off-by: Piotr Żelasko <petezor@gmail.com> * fix max_tps/max_tpt config parsing for list values Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> * Transition: tarred_random_access -> skip_missing_manifest_entries Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Improve warning messages Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Support single unified BPE tokenizer for Canary2 Signed-off-by: Piotr Zelasko <pzelasko@cs-oci-ord-login-01.cm.cluster> * bugfix +es to unified tokenizer support for Canary Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> * CanaryBPETokenizer Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Revert changes to fetching pad id in AED models Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> * Update docs Signed-off-by: Piotr Żelasko <petezor@gmail.com> * fix tests Signed-off-by: Piotr Żelasko <petezor@gmail.com> * Remove max_tps and max_duration from OOMptimizer output Signed-off-by: Piotr Żelasko <petezor@gmail.com> --------- Signed-off-by: Piotr Żelasko <petezor@gmail.com> Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com> Signed-off-by: pzelasko <pzelasko@users.noreply.github.com> Signed-off-by: Piotr Zelasko <pzelasko@cs-oci-ord-login-01.cm.cluster> Co-authored-by: pzelasko <pzelasko@users.noreply.github.com> Signed-off-by: Youngeun Kwon <youngeunk@nvidia.com>
What does this PR do ?
Improves the UX of estimating 2D buckets and preparing the training configuration. Key changes:
model.train_ds.bucketing_2d_strict_mode=False): find any bucket that will fit a given example. That means token-per-second outliers are pushed to buckets with higher durations increasing the padding but reducing the amount of discarded data. Use at your own risk - for the setups I tested it so far with, it may cause training instability (likely due to inclusion of lower-quality data).estimate_duration_bins_2d.pyscript.estimate_duration_bins_2d.py.tarred_random_accesstoskip_missing_manifest_entriesand adjusted the logic to reduce CPU memory usage and accelerate loading time.Collection: ASR, TTS, SpeechLLM
Changelog
Usage
# Add a code snippet demonstrating how to use thisGitHub 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:
PR Type:
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