Support whisper large/large-v1/large-v2/large-v3 and distil-large-v2 (#1114)

.github/workflows/export-whisper-to-onnx.yaml

@@ -15,9 +15,9 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-latest]
-        # model: ["distil-medium.en", "distil-small.en",  "tiny.en", "base.en", "small.en", "medium.en", "tiny", "base", "small", "medium", "large", "large-v1", "large-v2", "distil-large-v2"]
-        model: ["distil-medium.en", "distil-small.en",  "tiny.en", "base.en", "small.en", "medium.en", "tiny", "base", "small", "medium", "medium-aishell"]
+        os: [macos-latest]
+        model: ["distil-medium.en", "distil-small.en",  "tiny.en", "base.en", "small.en", "medium.en", "tiny", "base", "small", "medium", "medium-aishell", "large", "large-v1", "large-v2", "distil-large-v2"]
+        # model: ["large", "large-v1", "large-v2", "large-v3", "distil-large-v2"]
         python-version: ["3.8"]
 
     steps:
@@ -32,7 +32,7 @@ jobs:
         shell: bash
         run: |
           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
+          python3 -m pip install openai-whisper==20231117 onnxruntime onnx soundfile librosa
 
       - name: export ${{ matrix.model }}
         shell: bash
@@ -62,7 +62,6 @@ jobs:
             rm -fv medium-aishell-decoder.onnx
           fi
 
-
           ls -lh
 
           ls -lh ~/.cache/whisper || true
@@ -74,7 +73,8 @@ jobs:
           src=sherpa-onnx-whisper-${{ matrix.model }}
 
           cd ..
-          mv whisper $src
+          mkdir $src
+          mv -v whisper/$model* $src/
 
           echo "------------------------------"
 
@@ -97,19 +97,16 @@ jobs:
           ls -lh $src
           echo "--------------------"
 
-          if [[ $model == large || $model == large-v1 || $model == large-v2 || $model == distil-large-v2 ]]; then
-            #tar cvjf - $src | split --bytes=1024MB - $src.tar.bz2.
-            tar cvjf $src.tar.bz2 $src
-            split -b 1G $src.tar.bz2 $src.tar.bz2.
-            rm $src.tar.bz2
-            # cat $src.tar.gz.* | tar xjf -
+          if [[ $model == large || $model == large-v1 || $model == large-v2 || $model == distil-large-v2  ]]; then
+            echo "Don't release model to github for large models. $model"
           else
             tar cvjf $src.tar.bz2 $src
           fi
+
           ls -lh
 
-
       - name: Release
+        if: matrix.model != 'large' && matrix.model != 'large-v1' && matrix.model != 'large-v2' && matrix.model != 'large-v3' && matrix.model != 'distil-large-v2'
         uses: svenstaro/upload-release-action@v2
         with:
           file_glob: true
@@ -119,19 +116,6 @@ 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:
@@ -144,27 +128,36 @@ jobs:
 
           export GIT_CLONE_PROTECTION_ACTIVE=false
 
-          GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/csukuangfj/sherpa-onnx-whisper-${{ matrix.model }} huggingface
+          export GIT_LFS_SKIP_SMUDGE=1
+
+          git clone https://huggingface.co/csukuangfj/sherpa-onnx-whisper-${{ matrix.model }} huggingface
 
           if [[ $model != medium-aishell ]]; then
             rm -rf huggingface/*
           fi
 
-          if [[ $model == large || $model == large-v1 || $model == large-v2 || $model == distil-large-v2 ]]; then
-            mv $src.tar* ./huggingface
-          else
-            cp -v $src/*.onnx ./huggingface
-            cp -v $src/*tokens* ./huggingface
-            cp -av $src/test_wavs ./huggingface
-          fi
+          cp -av $src/* ./huggingface/
 
           cd huggingface
 
           git status
           ls -lh
-          git lfs track "*gz*"
           git lfs track "*onnx*"
+          git lfs track "*weights*"
 
           git add .
           git commit -m "upload ${{ matrix.model }}"
           git push https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/sherpa-onnx-whisper-${{ matrix.model }} main
+
+      - name: Test ${{ matrix.model }}
+        shell: bash
+        run: |
+          python3 -m pip install kaldi-native-fbank
+          git checkout .
+          model=${{ matrix.model }}
+          src=sherpa-onnx-whisper-$model
+          time python3 scripts/whisper/test.py \
+            --encoder $src/$model-encoder.onnx \
+            --decoder $src/$model-decoder.onnx \
+            --tokens $src/$model-tokens.txt \
+            $src/test_wavs/0.wav