debug

Author: winskuo-quic

.ci/scripts/test_model.sh

@@ -154,6 +154,7 @@ test_model_with_qnn() {
   export LD_LIBRARY_PATH=$QNN_SDK_ROOT/lib/x86_64-linux-clang/
   export PYTHONPATH=$EXECUTORCH_ROOT/..
 
+  EXTRA_FLAGS=""
   if [[ "${MODEL_NAME}" == "dl3" ]]; then
     EXPORT_SCRIPT=deeplab_v3
   elif [[ "${MODEL_NAME}" == "mv3" ]]; then
@@ -168,6 +169,8 @@ test_model_with_qnn() {
     EXPORT_SCRIPT=torchvision_vit
   elif [[ "${MODEL_NAME}" == "mb" ]]; then
     EXPORT_SCRIPT=mobilebert_fine_tune
+    EXTRA_FLAGS="--num_epochs 1"
+    pip install scikit-learn
   elif [[ "${MODEL_NAME}" == "w2l" ]]; then
     EXPORT_SCRIPT=wav2letter
   elif [[ "${MODEL_NAME}" == "edsr" ]]; then
@@ -183,7 +186,7 @@ test_model_with_qnn() {
   # TODO(guangyang): Make QNN chipset matches the target device
   QNN_CHIPSET=SM8450
 
-  "${PYTHON_EXECUTABLE}" -m examples.qualcomm.scripts.${EXPORT_SCRIPT} -b ${CMAKE_OUTPUT_DIR} -m ${QNN_CHIPSET} --compile_only
+  "${PYTHON_EXECUTABLE}" -m examples.qualcomm.scripts.${EXPORT_SCRIPT} -b ${CMAKE_OUTPUT_DIR} -m ${QNN_CHIPSET} --compile_only $EXTRA_FLAGS
   EXPORTED_MODEL=$(find "./${EXPORT_SCRIPT}" -type f -name "${MODEL_NAME}*.pte" -print -quit)
 }
 

examples/qualcomm/oss_scripts/llama/runner/runner.cpp

@@ -34,7 +34,9 @@ namespace example {
 
 namespace {
 static constexpr auto kTopp = 0.9f;
-void printReport(const Runner::Stats& stats, const std::string& performance_output_path);
+void printReport(
+    const Runner::Stats& stats,
+    const std::string& performance_output_path);
 std::string statsToJsonString(const Runner::Stats& stats);
 } // namespace
 
@@ -463,7 +465,9 @@ Error Runner::generate(
 }
 
 namespace {
-void printReport(const Runner::Stats& stats, const std::string& performance_output_path) {
+void printReport(
+    const Runner::Stats& stats,
+    const std::string& performance_output_path) {
   printf("PyTorchObserver %s\n", statsToJsonString(stats).c_str());
 
   ET_LOG(
@@ -524,7 +528,7 @@ void printReport(const Runner::Stats& stats, const std::string& performance_outp
 
   // For now, we just print the total inference time for CI, can save more info
   // in future if needed.
-  
+
   std::ofstream outfile(performance_output_path.c_str());
   if (outfile.is_open()) {
     double num_tok = (stats.num_generated_tokens) /

examples/qualcomm/scripts/mobilebert_fine_tune.py

@@ -169,7 +169,7 @@ def get_fine_tuned_mobilebert(artifacts_dir, pretrained_weight, batch_size):
     dataset_train = TensorDataset(input_ids_train, attention_masks_train, labels_train)
     dataset_val = TensorDataset(input_ids_val, attention_masks_val, labels_val)
 
-    epochs = 1 # Revert this later
+    epochs = args.num_epochs
     dataloader_train = DataLoader(
         dataset_train,
         sampler=RandomSampler(dataset_train),
@@ -365,6 +365,13 @@ def calibrator(gm):
         default=None,
         type=str,
     )
+    
+    parser.add_argument(
+        "--num_epochs",
+        help="If no pretrained weights are provided, set number of epochs to train the model",
+        default=5,
+        type=int,
+    )
 
     parser.add_argument(
         "-F",

examples/qualcomm/scripts/wav2letter.py

@@ -134,9 +134,12 @@ def main(args):
 
     # retrieve dataset, will take some time to download
     data_num = 100
-    inputs, targets, input_list = get_dataset(
-        data_size=data_num, artifact_dir=args.artifact
-    )
+    if args.compile_only:
+        inputs = [(torch.rand(1, 1, 454560, 1),)]
+    else:
+        inputs, targets, input_list = get_dataset(
+            data_size=data_num, artifact_dir=args.artifact
+        )
     pte_filename = "w2l_qnn"
     build_executorch_binary(
         model,