[WIP] Add NeMo Conformer TDT ASR support #1571
ysdede
wants to merge
42 commits into
huggingface:main
from
ysdede:v4-nemo-conformer-tdt-main-r3
feat(nemo-conformer-tdt): port Nemo Conformer TDT model and ASR pipeline
d0176023
fix(nemo-conformer-tdt): handle empty token decode output
964bc8fa
chore(nemo-conformer-tdt): keep typegen compatibility for transcribe/…
fa9bc254
refactor(nemo-conformer-tdt): redesign transcribe output shape and API
63aeee83
fix(nemo-conformer-tdt): round timestamps and confidences, simplify p…
f6835ad8
fix: dispose tensors on error path, decouple frame confidences from t…
2dd36a11
merge: sync with upstream (ModelRegistry API refactor, commit 4811a61)
10977df7
feat: integrate NemoConformerTDT with ModelRegistry API
39d9be4f
style: replace console.warn with logger.warn in feature extractor
3d984e52
fix(nemo-conformer-tdt): harden edge cases, restore pipeline design
9f3a2202
test(nemo-conformer-tdt): rewrite tests to match current API
3bac1dc0
fix(nemo-conformer-tdt): harden decoding and feature utilities
c75ebd27
fix(nemo-conformer-tdt): address critical review issues
493a5881
fix(nemo-conformer-tdt): clamp timestamps and validate cache limits
5b4cdabd
fix(nemo-conformer-tdt): close remaining tensor disposal leaks
76902274
fix(nemo-conformer-tdt): dispose auxiliary decoder outputs
1f065c36
perf(nemo-conformer-tdt): avoid tolist in length hot path
ec09a090
fix(nemo-conformer-tdt): harden duration and audio validation
8a90a7c9
fix: address prioritized review findings
ce0a3eb9
fix(nemo-conformer-tdt): apply low-risk hardening follow-ups
5d91d396
fix(nemo-conformer-tdt): enforce named outputs and frame-level confid…
dfc2c130
docs(nemo-conformer-tdt): clarify cached tensor sharing semantics
a5bd2cf7
fix(nemo-conformer-tdt): harden decoder I/O validation and feed cleanup
abada622
fix(nemo-conformer-tdt): address bot review findings
62d8bc0d
style(nemo-conformer-tdt): simplify duration frame expression
03fb8bd2
fix(nemo-tdt): address PR10 follow-up review comments
426061e5
fix(transformers): resolve Nemo TDT typegen regressions
d7476a6d
fix(nemo-tdt): address PR11 cache and vocab review feedback
0989f7ab
merge: sync upstream/main through #1559 and keep Nemo TDT ASR wiring
49a4af8f
fix(nemo-tdt): add supports() for ASR model class selection
ee819a1c
fix(model-registry): include processor files for text-to-audio pipelines
b44f7f3f
Revert "fix(model-registry): include processor files for text-to-audi…
bfa97e6e
fix(nemo-tdt): address PR #12 reviewer feedback
a85dff25
feat(nemo-tdt): align asr pipeline outputs and long-audio handling
8dfccddc
chore(tests): drop unrelated parakeet feature extractor coverage
816f5811
feat(nemo-conformer-tdt): add sentence-based ASR pipeline chunking
f59ba068
fix(nemo): scope ASR tests and address review fixes
00b3d934
fix(nemo-tdt): address follow-up review threads
07118c38
chore(asr): restore upstream cast spacing
341df3d7
fix(nemo-tdt): handle sparse vocab and merge dedupe
29f2baaf
fix(nemo-tdt): merge split domain suffix words
39e5cb1d
Revert "fix(nemo-tdt): merge split domain suffix words"
495bab59
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub