make builder accessible

examples/models/llama2/TARGETS

@@ -79,6 +79,7 @@ runtime.python_library(
         "//bento/...",
         "//bento_kernels/...",
         "//executorch/examples/...",
+        "@EXECUTORCH_CLIENTS",
     ],
     deps = [
         "//caffe2:torch",

examples/models/llama2/builder.py

@@ -156,6 +156,7 @@ def __init__(
         self.edge_manager: Optional[EdgeProgramManager] = None
         self.export_program = None
         self.output_dir = "."
+        self._saved_pte_filename = None
 
     def set_metadata(self, metadata: Optional[dict]) -> "LlamaEdgeManager":
         """
@@ -388,4 +389,11 @@ def save_to_pte(self, output_name: str) -> None:
             output_name (Optional[str]): The name of the .pte file.
         """
         assert output_name, "Need a valid output name"
-        save_pte_program(self.export_program, output_name, self.output_dir)
+        filename = save_pte_program(self.export_program, output_name, self.output_dir)
+        self._saved_pte_filename = filename
+
+    def get_saved_pte_filename(self) -> Optional[str]:
+        """
+        Return the filename of the most recenet saved .pte file. Return None if the model is not saved.
+        """
+        return self._saved_pte_filename

examples/models/llama2/export_llama_lib.py

@@ -293,14 +293,22 @@ def export_llama(modelname, args) -> str:
             from executorch.util.python_profiler import CProfilerFlameGraph
 
             with CProfilerFlameGraph(args.profile_path):
-                return _export_llama(modelname, args)
+                builder = _export_llama(modelname, args)
+                assert (
+                    filename := builder.get_saved_pte_filename()
+                ) is not None, "Fail to get file name from builder"
+                return filename
         except ImportError:
             print(
                 "Please run `pip install snakeviz` to install required dependencies for cProfiler flamegraph."
             )
             return ""
     else:
-        return _export_llama(modelname, args)
+        builder = _export_llama(modelname, args)
+        assert (
+            filename := builder.get_saved_pte_filename()
+        ) is not None, "Fail to get file name from builder"
+        return filename
 
 
 def _prepare_for_llama_export(modelname: str, args) -> LlamaEdgeManager:
@@ -383,7 +391,7 @@ def get_quantizer_and_quant_params(args):
     return pt2e_quant_params, quantizers, quant_dtype
 
 
-def _export_llama(modelname, args) -> str:  # noqa: C901
+def _export_llama(modelname, args) -> LlamaEdgeManager:  # noqa: C901
     pt2e_quant_params, quantizers, quant_dtype = get_quantizer_and_quant_params(args)
 
     # export_to_edge
@@ -468,4 +476,4 @@ def _export_llama(modelname, args) -> str:  # noqa: C901
 
     builder.save_to_pte(output_file)
 
-    return output_file
+    return builder

examples/portable/utils.py

@@ -102,7 +102,7 @@ def export_to_exec_prog(
 
 def save_pte_program(
     prog: ExecutorchProgramManager, model_name: str, output_dir: str = ""
-) -> None:
+) -> str:
     if model_name.endswith(".pte"):
         filename = model_name
     else:
@@ -114,3 +114,5 @@ def save_pte_program(
             logging.info(f"Saved exported program to {filename}")
     except Exception as e:
         logging.error(f"Error while saving to {filename}: {e}")
+
+    return filename