EngineX-Iluvatar/enginex-mr_series-sherpa-onnx
Archived
16
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Refactor exporting NeMo models (#2362)

Browse Source 
Refactors and extends model export support to include new NeMo Parakeet TDT int8 variants for English and Japanese, updating the Kotlin API, export scripts, test runners, and CI workflows.

- Added support for two new int8 model types in OfflineRecognizer.kt.
- Enhanced Python export scripts to perform dynamic quantization and metadata injection.
- Updated shell scripts and GitHub workflows to package, test, and publish int8 model artifacts.
This commit is contained in:
Fangjun Kuang
2025-07-09 16:02:12 +08:00
committed by GitHub
parent f1405779cf
commit 6122a678f5
19 changed files with 671 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

77
scripts/nemo/fast-conformer-hybrid-transducer-ctc/run-ctc-non-streaming.sh
View File

@@ -19,6 +19,12 @@ log "Process $name at $url"
d=sherpa-onnx-nemo-parakeet_tdt_ctc_110m-en-36000
mkdir -p $d
mv -v model.onnx $d/
cp -v tokens.txt $d/
ls -lh $d
d=sherpa-onnx-nemo-parakeet_tdt_ctc_110m-en-36000-int8
mkdir -p $d
mv -v model.int8.onnx $d/
mv -v tokens.txt $d/
ls -lh $d
@@ -33,6 +39,12 @@ log "Process $name at $url"
d=sherpa-onnx-nemo-fast-conformer-ctc-en-24500
mkdir -p $d
mv -v model.onnx $d/
cp -v tokens.txt $d/
ls -lh $d
d=sherpa-onnx-nemo-fast-conformer-ctc-en-24500-int8
mkdir -p $d
mv -v model.int8.onnx $d/
mv -v tokens.txt $d/
ls -lh $d
@@ -45,6 +57,12 @@ doc="This collection contains the Spanish FastConformer Hybrid (CTC and Transduc
d=sherpa-onnx-nemo-fast-conformer-ctc-es-1424
mkdir -p $d
mv -v model.onnx $d/
cp -v tokens.txt $d/
ls -lh $d
d=sherpa-onnx-nemo-fast-conformer-ctc-es-1424-int8
mkdir -p $d
mv -v model.int8.onnx $d/
mv -v tokens.txt $d/
ls -lh $d
@@ -57,6 +75,12 @@ doc="This collection contains the Multilingual FastConformer Hybrid (Transducer
d=sherpa-onnx-nemo-fast-conformer-ctc-en-de-es-fr-14288
mkdir -p $d
mv -v model.onnx $d/
cp -v tokens.txt $d/
ls -lh $d
d=sherpa-onnx-nemo-fast-conformer-ctc-en-de-es-fr-14288-int8
mkdir -p $d
mv -v model.int8.onnx $d/
mv -v tokens.txt $d/
ls -lh $d
@@ -69,6 +93,12 @@ doc="This collection contains the Multilingual FastConformer Hybrid (Transducer
d=sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k
mkdir -p $d
mv -v model.onnx $d/
cp -v tokens.txt $d/
ls -lh $d
d=sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k-int8
mkdir -p $d
mv -v model.int8.onnx $d/
mv -v tokens.txt $d/
ls -lh $d
@@ -92,6 +122,16 @@ mkdir -p $d/test_wavs
cp en.wav $d/test_wavs/0.wav
cp -v $data/en-english.wav $d/test_wavs/1.wav
d=sherpa-onnx-nemo-parakeet_tdt_ctc_110m-en-36000-int8
python3 ./test-onnx-ctc-non-streaming.py \
--model $d/model.int8.onnx \
--tokens $d/tokens.txt \
--wav $data/en-english.wav
mkdir -p $d/test_wavs
cp en.wav $d/test_wavs/0.wav
cp -v $data/en-english.wav $d/test_wavs/1.wav
d=sherpa-onnx-nemo-fast-conformer-ctc-en-24500
python3 ./test-onnx-ctc-non-streaming.py \
--model $d/model.onnx \
@@ -101,6 +141,15 @@ mkdir -p $d/test_wavs
cp en.wav $d/test_wavs/0.wav
cp -v $data/en-english.wav $d/test_wavs
d=sherpa-onnx-nemo-fast-conformer-ctc-en-24500-int8
python3 ./test-onnx-ctc-non-streaming.py \
--model $d/model.int8.onnx \
--tokens $d/tokens.txt \
--wav $data/en-english.wav
mkdir -p $d/test_wavs
cp en.wav $d/test_wavs/0.wav
cp -v $data/en-english.wav $d/test_wavs
d=sherpa-onnx-nemo-fast-conformer-ctc-es-1424
python3 ./test-onnx-ctc-non-streaming.py \
--model $d/model.onnx \
@@ -109,6 +158,14 @@ python3 ./test-onnx-ctc-non-streaming.py \
mkdir -p $d/test_wavs
cp -v $data/es-spanish.wav $d/test_wavs
d=sherpa-onnx-nemo-fast-conformer-ctc-es-1424-int8
python3 ./test-onnx-ctc-non-streaming.py \
--model $d/model.int8.onnx \
--tokens $d/tokens.txt \
--wav $data/es-spanish.wav
mkdir -p $d/test_wavs
cp -v $data/es-spanish.wav $d/test_wavs
d=sherpa-onnx-nemo-fast-conformer-ctc-en-de-es-fr-14288
mkdir -p $d/test_wavs
for w in en-english.wav de-german.wav es-spanish.wav fr-french.wav; do
@@ -119,6 +176,16 @@ for w in en-english.wav de-german.wav es-spanish.wav fr-french.wav; do
cp -v $data/$w $d/test_wavs
done
d=sherpa-onnx-nemo-fast-conformer-ctc-en-de-es-fr-14288-int8
mkdir -p $d/test_wavs
for w in en-english.wav de-german.wav es-spanish.wav fr-french.wav; do
python3 ./test-onnx-ctc-non-streaming.py \
--model $d/model.int8.onnx \
--tokens $d/tokens.txt \
--wav $data/$w
cp -v $data/$w $d/test_wavs
done
d=sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k
mkdir -p $d/test_wavs
for w in en-english.wav de-german.wav es-spanish.wav fr-french.wav hr-croatian.wav it-italian.wav po-polish.wav ru-russian.wav uk-ukrainian.wav; do
@@ -128,3 +195,13 @@ for w in en-english.wav de-german.wav es-spanish.wav fr-french.wav hr-croatian.w
--wav $data/$w
cp -v $data/$w $d/test_wavs
done
d=sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k-int8
mkdir -p $d/test_wavs
for w in en-english.wav de-german.wav es-spanish.wav fr-french.wav hr-croatian.wav it-italian.wav po-polish.wav ru-russian.wav uk-ukrainian.wav; do
python3 ./test-onnx-ctc-non-streaming.py \
--model $d/model.int8.onnx \
--tokens $d/tokens.txt \
--wav $data/$w
cp -v $data/$w $d/test_wavs
done