Skip to content

Commit 415b8c2

Browse files
csukuangfjcsukuangfj
authored
Update nemotron-speech-streaming-en-0.6b (k2-fsa#3555)
1 parent d795ccd commit 415b8c2

5 files changed

Lines changed: 295 additions & 196 deletions

File tree

‎.github/workflows/export-nemotron-speech-streaming-en-0.6b.yaml‎

Lines changed: 55 additions & 109 deletions
Original file line numberDiff line numberDiff line change
@@ -33,7 +33,8 @@ jobs:
3333
shell: bash
3434
run: |
3535
BRANCH='main'
36-
pip install git+https://github.com/NVIDIA/NeMo.git@$BRANCH#egg=nemo_toolkit[asr]
36+
# pip install git+https://github.com/NVIDIA/NeMo.git@$BRANCH#egg=nemo_toolkit[asr]
37+
pip install 'nemo-toolkit[asr]'
3738
pip install onnxruntime ipython
3839
pip install kaldi-native-fbank
3940
pip install soundfile librosa
@@ -45,46 +46,54 @@ jobs:
4546
4647
python3 ./export_onnx.py
4748
48-
ls -lh *.onnx
49+
ls -lh
50+
echo "---"
51+
52+
ls -lh */
53+
4954
echo "---"
50-
ls -lh encoder.*
5155
5256
- name: Collect results
5357
shell: bash
5458
run: |
5559
src=scripts/nemo/nemotron-speech-streaming-en-0.6b
56-
d=sherpa-onnx-nemotron-speech-streaming-en-0.6b-2026-01-14
57-
mkdir -p $d
58-
59-
cp -av $src/encoder.onnx $d/
60-
cp -av $src/encoder.data $d/
61-
cp -av $src/decoder.onnx $d/
62-
cp -av $src/joiner.onnx $d/
63-
cp -av $src/tokens.txt $d/
64-
cat >$d/README.md <<EOF
65-
# Introduction
66-
This model is from https://huggingface.co/nvidia/nemotron-speech-streaming-en-0.6b
67-
EOF
6860
69-
ls -lh $d
61+
for chunk in 80 160 560 1120; do
62+
d=sherpa-onnx-nemotron-speech-streaming-en-0.6b-${chunk}ms-2026-04-25
63+
mkdir -p $d
7064
71-
d=sherpa-onnx-nemotron-speech-streaming-en-0.6b-int8-2026-01-14
72-
mkdir -p $d
65+
cp -av $src/$chunk/encoder.onnx $d/
66+
cp -av $src/$chunk/encoder.data $d/
67+
cp -av $src/$chunk/decoder.onnx $d/
68+
cp -av $src/$chunk/joiner.onnx $d/
69+
cp -av $src/tokens.txt $d/
70+
cat >$d/README.md <<EOF
71+
# Introduction
72+
This model is from https://huggingface.co/nvidia/nemotron-speech-streaming-en-0.6b
7373
74-
cp -av $src/encoder.int8.onnx $d/
75-
cp -av $src/decoder.int8.onnx $d/
76-
cp -av $src/joiner.int8.onnx $d/
77-
cp -av $src/tokens.txt $d/
78-
cat >$d/README.md <<EOF
79-
# Introduction
80-
This model is from https://huggingface.co/nvidia/nemotron-speech-streaming-en-0.6b
74+
The chunk size is ${chunk}ms for the exported ONNX model.
8175
EOF
8276
83-
ls -lh $d
77+
ls -lh $d
78+
79+
d=sherpa-onnx-nemotron-speech-streaming-en-0.6b-${chunk}ms-int8-2026-04-25
80+
mkdir -p $d
81+
82+
cp -av $src/${chunk}/encoder.int8.onnx $d/
83+
cp -av $src/${chunk}/decoder.int8.onnx $d/
84+
cp -av $src/${chunk}/joiner.int8.onnx $d/
85+
cp -av $src/tokens.txt $d/
86+
cat >$d/README.md <<EOF
87+
# Introduction
88+
This model is from https://huggingface.co/nvidia/nemotron-speech-streaming-en-0.6b
89+
90+
The chunk size is ${chunk}ms for the exported ONNX model.
91+
EOF
8492
93+
ls -lh $d
94+
done
8595
8696
- name: Download test waves
87-
if: true
8897
shell: bash
8998
run: |
9099
mkdir test_wavs
@@ -96,8 +105,14 @@ jobs:
96105
popd
97106
98107
models=(
99-
sherpa-onnx-nemotron-speech-streaming-en-0.6b-int8-2026-01-14
100-
sherpa-onnx-nemotron-speech-streaming-en-0.6b-2026-01-14
108+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-80ms-int8-2026-04-25
109+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-160ms-int8-2026-04-25
110+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-560ms-int8-2026-04-25
111+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-1120ms-int8-2026-04-25
112+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-80ms-2026-04-25
113+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-160ms-2026-04-25
114+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-560ms-2026-04-25
115+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-1120ms-2026-04-25
101116
)
102117
for m in ${models[@]}; do
103118
cp -av test_wavs $m
@@ -111,7 +126,7 @@ jobs:
111126
with:
112127
file_glob: true
113128
# fp32 models is 2.2GB > 2GB
114-
file: sherpa-onnx-nemotron-speech-streaming-en-0.6b-int8-2026-01-14.tar.bz2
129+
file: sherpa-onnx-nemotron-speech-streaming-en-0.6b-*-int8-2026-04-25.tar.bz2
115130
overwrite: true
116131
repo_name: k2-fsa/sherpa-onnx
117132
repo_token: ${{ secrets.UPLOAD_GH_SHERPA_ONNX_TOKEN }}
@@ -130,8 +145,14 @@ jobs:
130145
git config --global user.name "Fangjun Kuang"
131146
132147
models=(
133-
sherpa-onnx-nemotron-speech-streaming-en-0.6b-int8-2026-01-14
134-
sherpa-onnx-nemotron-speech-streaming-en-0.6b-2026-01-14
148+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-80ms-int8-2026-04-25
149+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-160ms-int8-2026-04-25
150+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-560ms-int8-2026-04-25
151+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-1120ms-int8-2026-04-25
152+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-80ms-2026-04-25
153+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-160ms-2026-04-25
154+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-560ms-2026-04-25
155+
sherpa-onnx-nemotron-speech-streaming-en-0.6b-1120ms-2026-04-25
135156
)
136157
137158
for m in ${models[@]}; do
@@ -143,7 +164,7 @@ jobs:
143164
rm -rf huggingface
144165
export GIT_LFS_SKIP_SMUDGE=1
145166
export GIT_CLONE_PROTECTION_ACTIVE=false
146-
git clone https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$m huggingface
167+
git clone https://csukuangfj2:$HF_TOKEN@huggingface.co/csukuangfj2/$m huggingface
147168
cp -av $m/* huggingface
148169
cd huggingface
149170
git lfs track "*.onnx"
@@ -153,83 +174,8 @@ jobs:
153174
git add .
154175
git status
155176
git commit -m "first commit"
156-
git push https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$m main
177+
git push https://csukuangfj2:$HF_TOKEN@huggingface.co/csukuangfj2/$m main
157178
cd ..
158179
done
159180
160181
rm -rf huggingface
161-
162-
- name: Publish to modelscope
163-
env:
164-
MS_TOKEN: ${{ secrets.MODEL_SCOPE_GIT_TOKEN }}
165-
uses: nick-fields/retry@v3
166-
with:
167-
max_attempts: 20
168-
timeout_seconds: 200
169-
shell: bash
170-
command: |
171-
git config --global user.email "csukuangfj@gmail.com"
172-
git config --global user.name "Fangjun Kuang"
173-
for m in "*.tar.bz2"; do
174-
export GIT_LFS_SKIP_SMUDGE=1
175-
export GIT_CLONE_PROTECTION_ACTIVE=false
176-
177-
rm -rf ms
178-
git clone https://oauth2:${MS_TOKEN}@www.modelscope.cn/csukuangfj/asr-models.git ms
179-
180-
mkdir ms/nemo
181-
cp -av $m ms/nemo
182-
183-
pushd ms
184-
git lfs track "*.tar.bz2"
185-
git status
186-
ls -lh
187-
git add .
188-
189-
git commit -m "add models"
190-
git push https://oauth2:${MS_TOKEN}@www.modelscope.cn/csukuangfj/asr-models.git
191-
192-
popd
193-
done
194-
rm -rf ms
195-
196-
- name: Publish to huggingface
197-
env:
198-
HF_TOKEN: ${{ secrets.HF_TOKEN }}
199-
uses: nick-fields/retry@v3
200-
with:
201-
max_attempts: 20
202-
timeout_seconds: 200
203-
shell: bash
204-
command: |
205-
git config --global user.email "csukuangfj@gmail.com"
206-
git config --global user.name "Fangjun Kuang"
207-
for m in "*.tar.bz2"; do
208-
export GIT_LFS_SKIP_SMUDGE=1
209-
export GIT_CLONE_PROTECTION_ACTIVE=false
210-
rm -rf huggingface
211-
git clone https://csukuangfj:$HF_TOKEN@huggingface.co/k2-fsa/sherpa-onnx-models huggingface
212-
213-
d=asr-models/nemo
214-
mkdir -p huggingface/$d
215-
216-
cp -v $m huggingface/$d/
217-
218-
pushd huggingface
219-
git lfs track "*.tar.bz2"
220-
ls -lh $d/$m
221-
222-
ls -lh $d
223-
224-
pushd $d
225-
git lfs track "*.tar.bz2"
226-
popd
227-
228-
git status
229-
git add .
230-
231-
git commit -m "add $m"
232-
git push https://csukuangfj:$HF_TOKEN@huggingface.co/k2-fsa/sherpa-onnx-models main
233-
popd
234-
done
235-
rm -rf huggingface

‎scripts/apk/generate-asr-apk-script.py‎

Lines changed: 60 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -511,6 +511,66 @@ def get_models():
511511
512512
ls -lh
513513
514+
popd
515+
""",
516+
),
517+
Model(
518+
model_name="sherpa-onnx-nemotron-speech-streaming-en-0.6b-80ms-int8-2026-04-25",
519+
idx=30,
520+
lang="en",
521+
short_name="nemotron-speech-streaming-en-0.6b-80ms-int8-2026-04-25",
522+
cmd="""
523+
pushd $model_name
524+
525+
rm -rf test_wavs
526+
527+
ls -lh
528+
529+
popd
530+
""",
531+
),
532+
Model(
533+
model_name="sherpa-onnx-nemotron-speech-streaming-en-0.6b-160ms-int8-2026-04-25",
534+
idx=31,
535+
lang="en",
536+
short_name="nemotron-speech-streaming-en-0.6b-160ms-int8-2026-04-25",
537+
cmd="""
538+
pushd $model_name
539+
540+
rm -rf test_wavs
541+
542+
ls -lh
543+
544+
popd
545+
""",
546+
),
547+
Model(
548+
model_name="sherpa-onnx-nemotron-speech-streaming-en-0.6b-560ms-int8-2026-04-25",
549+
idx=32,
550+
lang="en",
551+
short_name="nemotron-speech-streaming-en-0.6b-560ms-int8-2026-04-25",
552+
cmd="""
553+
pushd $model_name
554+
555+
rm -rf test_wavs
556+
557+
ls -lh
558+
559+
popd
560+
""",
561+
),
562+
Model(
563+
model_name="sherpa-onnx-nemotron-speech-streaming-en-0.6b-1120ms-int8-2026-04-25",
564+
idx=33,
565+
lang="en",
566+
short_name="nemotron-speech-streaming-en-0.6b-1120ms-int8-2026-04-25",
567+
cmd="""
568+
pushd $model_name
569+
570+
rm -rf test_wavs
571+
572+
ls -lh
573+
514574
popd
515575
""",
516576
),

0 commit comments

Comments
 (0)