Fix reporting backends and dtyep to benchmark results

Author: Guang Yang

.ci/scripts/test_llama.sh

@@ -171,7 +171,7 @@ else
 fi
 
 # Check dtype.
-EXPORTED_MODEL_NAME="llama2"
+EXPORTED_MODEL_NAME="tinyllama_${MODE}_${DTYPE}"
 if [[ "${DTYPE}" == "fp16" ]]; then
   EXPORTED_MODEL_NAME="${EXPORTED_MODEL_NAME}_h"
 elif [[ "${DTYPE}" == "bf16" ]]; then

.ci/scripts/test_model.sh

@@ -152,33 +152,35 @@ test_model_with_qnn() {
 
   export LD_LIBRARY_PATH=$QNN_SDK_ROOT/lib/x86_64-linux-clang/
   export PYTHONPATH=$EXECUTORCH_ROOT/..
+  # QNN support fp16 only
+  DTYPE=fp16
 
   if [[ "${MODEL_NAME}" == "dl3" ]]; then
     EXPORT_SCRIPT=deeplab_v3
-    EXPORTED_MODEL_NAME=dlv3_qnn.pte
+    EXPORTED_MODEL_NAME=deeplab-v3_qnn_${DTYPE}.pte
   elif [[ "${MODEL_NAME}" == "mv3" ]]; then
     EXPORT_SCRIPT=mobilenet_v3
-    EXPORTED_MODEL_NAME=mv3_qnn.pte
+    EXPORTED_MODEL_NAME=mobilenet-v3_qnn_${DTYPE}.pte
   elif [[ "${MODEL_NAME}" == "mv2" ]]; then
     EXPORT_SCRIPT=mobilenet_v2
-    EXPORTED_MODEL_NAME=mv2_qnn.pte
+    EXPORTED_MODEL_NAME=mobilenet-v2_qnn_${DTYPE}.pte
   elif [[ "${MODEL_NAME}" == "ic4" ]]; then
     EXPORT_SCRIPT=inception_v4
-    EXPORTED_MODEL_NAME=ic4_qnn.pte
+    EXPORTED_MODEL_NAME=inception-v4_qnn_${DTYPE}.pte
   elif [[ "${MODEL_NAME}" == "ic3" ]]; then
     EXPORT_SCRIPT=inception_v3
-    EXPORTED_MODEL_NAME=ic3_qnn.pte
+    EXPORTED_MODEL_NAME=inception-v3_qnn_${DTYPE}.pte
   elif [[ "${MODEL_NAME}" == "vit" ]]; then
     EXPORT_SCRIPT=torchvision_vit
-    EXPORTED_MODEL_NAME=vit_qnn.pte
+    EXPORTED_MODEL_NAME=torchvision-vit_qnn_${DTYPE}.pte
   fi
 
   # Use SM8450 for S22, SM8550 for S23, and SM8560 for S24
   # 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 ${QNN_CHIPSET} --compile_only
-  EXPORTED_MODEL=./${EXPORT_SCRIPT}/${EXPORTED_MODEL_NAME}
+  "${PYTHON_EXECUTABLE}" -m examples.qualcomm.scripts.${EXPORT_SCRIPT} -b ${CMAKE_OUTPUT_DIR} -m ${QNN_CHIPSET} -a ${EXPORTED_MODEL_NAME} --compile_only
+  EXPORTED_MODEL=./${EXPORTED_MODEL_NAME}
 }
 
 test_model_with_coreml() {

examples/apple/coreml/scripts/export.py

@@ -110,9 +110,9 @@ def export_lowered_module_to_executorch_program(lowered_module, example_inputs):
     return exec_prog
 
 
-def save_executorch_program(exec_prog, model_name, compute_unit):
+def save_executorch_program(exec_prog, model_name, compute_unit, precision):
     buffer = exec_prog.buffer
-    filename = f"{model_name}_coreml_{compute_unit}.pte"
+    filename = f"{model_name}_coreml_{compute_unit}_{precision}.pte"
     print(f"Saving exported program to {filename}")
     with open(filename, "wb") as file:
         file.write(buffer)
@@ -189,7 +189,9 @@ def main():
             example_inputs,
         )
 
-    save_executorch_program(exec_program, args.model_name, args.compute_unit)
+    save_executorch_program(
+        exec_program, args.model_name, args.compute_unit, args.compute_precision
+    )
     generate_etrecord(f"{args.model_name}_coreml_etrecord.bin", edge_copy, exec_program)
 
     if args.save_processed_bytes and lowered_module is not None: