Fix whisper test script for the latest onnxruntime. (#494)

k2-fsa · Dec 20, 2023 · ef8d112 · ef8d112
1 parent 03ff9db
commit ef8d112
Show file tree

Hide file tree

Showing 3 changed files with 18 additions and 3 deletions.
diff --git a/.github/workflows/export-whisper-to-onnx.yaml b/.github/workflows/export-whisper-to-onnx.yaml
@@ -31,7 +31,7 @@ jobs:
       - name: Install dependencies
         shell: bash
         run: |
-          python3 -m pip install torch==1.13.0 -f https://download.pytorch.org/whl/cpu/torch_stable.html
+          python3 -m pip install torch==1.13.0 torchaudio==0.13.0 -f https://download.pytorch.org/whl/cpu/torch_stable.html
           python3 -m pip install openai-whisper==20230314 onnxruntime onnx
 
       - name: export ${{ matrix.model }}
@@ -108,6 +108,19 @@ jobs:
           repo_token: ${{ secrets.UPLOAD_GH_SHERPA_ONNX_TOKEN }}
           tag: asr-models
 
+      - name: Test ${{ matrix.model }}
+        shell: bash
+        run: |
+          python3 -m pip install kaldi-native-fbank
+          git checkout .
+          model=${{ matrix.model }}
+          src=sherpa-onnx-whisper-$model
+          python3 scripts/whisper/test.py \
+            --encoder $src/$model-encoder.int8.onnx \
+            --decoder $src/$model-decoder.int8.onnx \
+            --tokens $src/$model-tokens.txt \
+            $src/test_wavs/0.wav
+
       - name: Publish ${{ matrix.model }} to huggingface
         shell: bash
         env:

diff --git a/build-apk-two-pass.sh b/build-apk-two-pass.sh
@@ -74,11 +74,11 @@ git lfs pull --include "*.onnx"
 
 # remove .git to save spaces
 rm -rf .git
-rm README.md
+rm -fv README.md
 rm -rf test_wavs
 rm .gitattributes
 
-rm *.ort
+rm -fv *.ort
 rm tiny.en-encoder.onnx
 rm tiny.en-decoder.onnx
 

diff --git a/scripts/whisper/test.py b/scripts/whisper/test.py
@@ -82,6 +82,7 @@ def init_encoder(self, encoder: str):
         self.encoder = ort.InferenceSession(
             encoder,
             sess_options=self.session_opts,
+            providers=["CPUExecutionProvider"],
         )
 
         meta = self.encoder.get_modelmeta().custom_metadata_map
@@ -113,6 +114,7 @@ def init_decoder(self, decoder: str):
         self.decoder = ort.InferenceSession(
             decoder,
             sess_options=self.session_opts,
+            providers=["CPUExecutionProvider"],
         )
 
     def run_encoder(