remove args

Author: angelayi

backends/qualcomm/qnn_preprocess.py

@@ -67,6 +67,8 @@ def preprocess(
         for node in pass_result.graph_module.graph.nodes:
             if node.op == "call_function":
                 logger.info(f"Visiting: {node}, {node.target.__name__}")
+                if node.target.__name__ == "aten.index_put.default":
+                    logger.info("lfq: node ", node.args)
                 if node.target.__name__ in node_visitors:
                     py_op_wrapper = node_visitors[node.target.__name__].define_node(
                         node, nodes_to_wrappers

examples/models/llama/export_llama_lib.py

@@ -663,6 +663,12 @@ def _export_llama(args) -> LLMEdgeManager:  # noqa: C901
         quantizers
     ).export_to_edge()
 
+    accuracy = torch.allclose(
+        builder_exported_to_edge.edge_manager.exported_program()(*self.example_inputs),
+        builder_exported_to_edge.model(*self.example_inputs),
+    )
+    print(f"lfq: post to_edge accuracy: {accuracy}")
+
     modelname = builder_exported_to_edge.modelname
 
     # to_backend
@@ -800,6 +806,11 @@ def _export_llama(args) -> LLMEdgeManager:  # noqa: C901
             # pyre-fixme[16]: Module `backends` has no attribute `qualcomm`.
             canonicalize_program(builder.edge_manager.exported_program())
 
+        print("lfq: exported program after to_backend, graph_module")
+        print(builder.edge_manager.exported_program().graph_module)
+        print("lfq: exported program after to_backend, print_readable")
+        print(builder.edge_manager.exported_program().graph_module.print_readable())
+
         builder = builder.to_executorch()
 
     if args.profile_memory:

examples/models/llama3_2_vision/text_decoder/test/test_text_decoder.py

@@ -74,8 +74,6 @@ def test_llama3_2_text_decoder_aoti(self) -> None:
         with tempfile.TemporaryDirectory() as tmpdir:
             path = torch._inductor.aoti_compile_and_package(
                 ep,
-                model.get_example_inputs(),
-                kwargs=model.get_example_kwarg_inputs(),
                 package_path=os.path.join(tmpdir, "text_decoder.pt2"),
             )
             encoder_aoti = torch._inductor.aoti_load_package(path)

examples/models/llama3_2_vision/vision_encoder/test/test_vision_encoder.py

@@ -36,7 +36,6 @@ def test_flamingo_vision_encoder(self) -> None:
         with tempfile.TemporaryDirectory() as tmpdir:
             path = torch._inductor.aoti_compile_and_package(
                 ep,
-                model.get_example_inputs(),
                 package_path=os.path.join(tmpdir, "vision_encoder.pt2"),
             )
             print(path)

extension/llm/export/builder.py

@@ -193,6 +193,19 @@ def export(self) -> "LLMEdgeManager":
                     dynamic_shapes=dynamic_shape,
                     strict=True,
                 )
+                print("lfq: pre exported_module ", exported_module.graph_module)
+                print("lfq: pre exported graph ", exported_module.graph)
+
+                accuracy = torch.allclose(
+                    exported_module.module()(*self.example_inputs),
+                    self.model(*self.example_inputs),
+                )
+                print("lfq: pre to_edge accuracy ", accuracy)
+
+                exported_module = exported_module.run_decompositions({})
+                print("lfq: post exported_module ", exported_module.graph_module)
+                print("lfq: post exported graph ", exported_module.graph)
+
             else:
                 logging.info("Exporting with:")
                 logging.info(f"inputs: {self.example_inputs}")

extension/llm/modules/test/test_position_embeddings.py

@@ -177,7 +177,6 @@ def test_tiled_token_positional_embedding_aoti(self):
         with tempfile.TemporaryDirectory() as tmpdir:
             path = torch._inductor.aoti_compile_and_package(
                 tpe_ep,
-                (self.x, self.aspect_ratio),
                 package_path=os.path.join(tmpdir, "tpe.pt2"),
             )
             tpe_aoti = load_package(path)