rebase

Author: winskuo-quic

.ci/scripts/test_model.sh

@@ -188,6 +188,12 @@ test_model_with_qnn() {
     EXPORT_SCRIPT=edsr
     # Additional deps for edsr
     pip install piq
+  elif [[ "${MODEL_NAME}" == "deit" ]]; then
+    EXPORT_SCRIPT=deit
+  elif [[ "${MODEL_NAME}" == "pvt" ]]; then
+    EXPORT_SCRIPT=pvt
+  elif [[ "${MODEL_NAME}" == "swin" ]]; then
+    EXPORT_SCRIPT=swin_transformer
   elif [[ "${MODEL_NAME}" == "albert" ]]; then
     EXPORT_SCRIPT=albert
   elif [[ "${MODEL_NAME}" == "bert" ]]; then
@@ -196,26 +202,13 @@ test_model_with_qnn() {
     EXPORT_SCRIPT=distilbert
   elif [[ "${MODEL_NAME}" == "eurobert" ]]; then
     EXPORT_SCRIPT=eurobert
+  elif [[ "${MODEL_NAME}" == "roberta" ]]; then
+    EXPORT_SCRIPT=roberta
   else
     echo "Unsupported model $MODEL_NAME"
     exit 1
   fi
 
-  SCRIPT_FOLDER=""
-  case "${MODEL_NAME}" in
-    "dl3"|"mv3"|"mv2"|"ic4"|"ic3"|"vit"|"mb"|"w2l")
-        SCRIPT_FOLDER=scripts
-        ;;
-    "deit"|"pvt"|"swin")
-        SCRIPT_FOLDER=oss_scripts
-        ;;
-    *)
-        echo "Unsupported model $MODEL_NAME"
-        exit 1
-        ;;
-  esac
-  
-
   # Use SM8450 for S22, SM8550 for S23, and SM8560 for S24
   # TODO(guangyang): Make QNN chipset matches the target device
   QNN_CHIPSET=SM8450
@@ -225,7 +218,10 @@ test_model_with_qnn() {
     "dl3"|"mv3"|"mv2"|"ic4"|"ic3"|"vit"|"mb"|"w2l")
         SCRIPT_FOLDER=scripts
         ;;
-    "albert"|"bert"|"distilbert")
+    "deit"|"pvt"|"swin")
+        SCRIPT_FOLDER=oss_scripts
+        ;;
+    "albert"|"bert"|"distilbert"|"roberta")
         pip install evaluate
         SCRIPT_FOLDER=oss_scripts
         # Bert models running in 16bit will encounter op validation fail on some operations,

.github/workflows/trunk.yml

@@ -479,32 +479,6 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-qnn-deps.sh
         PYTHON_EXECUTABLE=python bash .ci/scripts/build-qnn-sdk.sh
         PYTHON_EXECUTABLE=python bash .ci/scripts/test_model.sh ${{ matrix.model }} "cmake" "qnn"
-  
-  test-qnn-optimum-model:
-    name: test-qnn-optimum-model
-    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
-    permissions:
-      id-token: write
-      contents: read
-    strategy:
-      matrix:
-        dtype: [fp32]
-        model: [deit, pvt, swin]
-      fail-fast: false
-    with:
-      runner: linux.2xlarge
-      docker-image: executorch-ubuntu-22.04-qnn-sdk
-      submodules: 'recursive'
-      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
-      timeout: 900
-      script: |
-        # The generic Linux job chooses to use base env, not the one setup by the image
-        CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
-        conda activate "${CONDA_ENV}"
-        PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh --build-tool cmake
-        PYTHON_EXECUTABLE=python bash .ci/scripts/setup-qnn-deps.sh
-        PYTHON_EXECUTABLE=python bash .ci/scripts/build-qnn-sdk.sh
-        PYTHON_EXECUTABLE=python bash .ci/scripts/test_model.sh ${{ matrix.model }} "cmake" "qnn"
 
   test-qnn-optimum-model:
     name: test-qnn-optimum-model
@@ -515,7 +489,7 @@ jobs:
     strategy:
       matrix:
         dtype: [fp32]
-        model: [albert, bert, distilbert] # eurobert requires transfomer >= 4.48.0, skip for now
+        model: [deit, pvt, swin, albert, bert, distilbert, roberta] # eurobert requires transfomer >= 4.48.0, skip for now
       fail-fast: false
     with:
       runner: linux.2xlarge

examples/qualcomm/oss_scripts/roberta.py

@@ -47,7 +47,7 @@ def main(args):
 
     # Get the Roberta model.
     model = get_instance(args)
-    pte_filename = "roberta_qnn"
+    pte_filename = "roberta_qnn_q16"
 
     # lower to QNN
     passes_job = get_capture_program_passes()
@@ -137,7 +137,7 @@ def main(args):
         "-a",
         "--artifact",
         help="path for storing generated artifacts and output by this example. Default ./Roberta_qnn",
-        default="./Roberta_qnn",
+        default="./roberta",
         type=str,
     )
     parser.add_argument(