Merge branch 'main' into add-cast-int64-placeholders-to-int32-pass

Author: YufengShi-dudu

backends/qualcomm/tests/test_qnn_delegate.py

@@ -4979,9 +4979,9 @@ def test_static_qwen2_5(self):
             if "Error" in msg:
                 self.fail(msg["Error"])
             else:
-                inference_speed_ref = {"SM8650": 110, "SM8750": 130}
+                inference_speed_ref = {"SM8650": 115, "SM8750": 155}
                 self.assertLessEqual(msg["wiki_ppl"], 15)
-                self.assertLessEqual(msg["pte_size"], 800000000)  # 800mb
+                self.assertLessEqual(msg["pte_size"], 600000000)  # 600mb
                 if self.model in inference_speed_ref:
                     self.assertGreaterEqual(
                         msg["inference_speed"], inference_speed_ref[self.model]

backends/transforms/remove_clone_ops.py

@@ -22,6 +22,7 @@ class RemoveCloneOpsTransform(ExportPass):
 
     clone_ops: Set[torch._ops.OpOverload] = {
         exir_ops.edge.aten.clone.default,
+        exir_ops.edge.dim_order_ops._clone_dim_order.default,
     }
 
     def __init__(self) -> None:
@@ -34,12 +35,15 @@ def _remove(self, graph_module: torch.fx.GraphModule) -> None:
             if n.target not in self.clone_ops:
                 continue
 
-            to_be_remove = n
+            if self._is_non_identity_clone(n):
+                continue
+
+            to_be_removed = n
             for user_n in list(n.users.keys()):
                 user_n.replace_input_with(n, n.args[0])
             if n.args[0].target in _DEQUANT_OPS:
                 dequant_nodes += [n.args[0]]
-            graph_module.graph.erase_node(to_be_remove)
+            graph_module.graph.erase_node(to_be_removed)
 
         eliminate_dq_q(graph_module, dequant_nodes)
 
@@ -48,3 +52,27 @@ def call(self, graph_module: torch.fx.GraphModule) -> PassResult:
         graph_module.recompile()
         dead_code_elimination_pass(graph_module)
         return PassResult(graph_module, True)
+
+    def _is_non_identity_clone(self, node: torch.fx.Node) -> bool:
+        """Return True if clone has modified memory layout or dim order."""
+
+        # aten.clone: check for memory_format changes
+        if node.target == exir_ops.edge.aten.clone.default:
+            memory_format = node.kwargs.get("memory_format")
+            if memory_format in (None, torch.preserve_format):
+                return False
+            input_meta = node.args[0].meta
+            return "val" in input_meta and not input_meta["val"].is_contiguous(
+                memory_format=memory_format
+            )
+
+        # _clone_dim_order: check for dim_order changes
+        if node.target == exir_ops.edge.dim_order_ops._clone_dim_order.default:
+            input_meta = node.args[0].meta
+            return (
+                "val" in node.meta
+                and "val" in input_meta
+                and node.meta["val"].dim_order() != input_meta["val"].dim_order()
+            )
+
+        return False

backends/transforms/test/test_remove_clone_ops.py

@@ -8,13 +8,30 @@
 
 import torch
 from executorch.backends.transforms.remove_clone_ops import RemoveCloneOpsTransform
+from executorch.exir import EdgeCompileConfig, to_edge
 from executorch.exir.dialects._ops import ops as exir_ops
+from executorch.exir.dim_order_utils import is_channel_last_dim_order
+from executorch.exir.tests.test_memory_format_ops_pass_utils import (
+    SimpleCloneChannelsLastModule,
+)
+from torch.export import export
 from torch.fx import GraphModule
 from torch.testing import FileCheck
 from torch.testing._internal.common_utils import TestCase
 
 
 class TestRemoveCloneOpsTransform(TestCase):
+    # Clone ops can appear as either aten.clone or _clone_dim_order depending on the _skip_dim_order flag.
+    # _skip_dim_order=True tests aten.clone
+    # _skip_dim_order=False tests _clone_dim_order
+    CLONE_OP_CASES = [
+        (True, "executorch_exir_dialects_edge__ops_aten_clone_default"),
+        (
+            False,
+            "executorch_exir_dialects_edge__ops_dim_order_ops__clone_dim_order_default",
+        ),
+    ]
+
     def test_dq_clone_q_linear(self):
         """
         Test RemoveCloneOpsTransform on a graph with d/q -> clone -> q -> linear pattern
@@ -123,6 +140,58 @@ def forward(self, x):
             transformed_gm.code
         )
 
+    def test_clone_non_identity_survives(self):
+        """Verify clone ops that modify memory_format are preserved by RemoveCloneOpsTransform."""
+
+        for skip_dim_order, clone_op_str in self.CLONE_OP_CASES:
+            model = SimpleCloneChannelsLastModule()
+            x = torch.randn(3, 4, 5, 6).to(memory_format=torch.contiguous_format)
+
+            exported = export(model.eval(), (x,), strict=True)
+            before_epm = to_edge(
+                exported,
+                compile_config=EdgeCompileConfig(_skip_dim_order=skip_dim_order),
+            )
+
+            updated_epm = before_epm.transform([RemoveCloneOpsTransform()])
+
+            FileCheck().check_count(clone_op_str, 1, exactly=True).run(
+                updated_epm.exported_program().graph_module.code
+            )
+
+            expected = before_epm.exported_program().module()(x)
+            actual = updated_epm.exported_program().module()(x)
+            assert torch.allclose(actual, expected)
+            assert is_channel_last_dim_order(actual)
+
+    def test_clone_identity_removed(self):
+        """Verify identity clone ops are removed by RemoveCloneOpsTransform."""
+
+        for skip_dim_order, clone_op_str in self.CLONE_OP_CASES:
+            model = SimpleCloneChannelsLastModule()
+            x = torch.randn(3, 4, 5, 6).to(memory_format=torch.channels_last)
+
+            exported = export(model.eval(), (x,), strict=True)
+            before_epm = to_edge(
+                exported,
+                compile_config=EdgeCompileConfig(_skip_dim_order=skip_dim_order),
+            )
+
+            FileCheck().check_count(clone_op_str, 1, exactly=True).run(
+                before_epm.exported_program().graph_module.code
+            )
+
+            updated_epm = before_epm.transform([RemoveCloneOpsTransform()])
+
+            FileCheck().check_not(clone_op_str).run(
+                updated_epm.exported_program().graph_module.code
+            )
+
+            expected = before_epm.exported_program().module()(x)
+            actual = updated_epm.exported_program().module()(x)
+            assert torch.allclose(actual, expected)
+            assert is_channel_last_dim_order(actual)
+
 
 if __name__ == "__main__":
     unittest.main()

examples/qualcomm/oss_scripts/llama/__init__.py

@@ -211,8 +211,8 @@ class Qwen2_5_0_5B(LLMModelConfig):
 
     num_sharding = 1
     # quant config
-    ptq = QuantDtype.use_16a8w
-    group_size = None
+    ptq = QuantDtype.use_16a4w_block
+    group_size = 16
     masked_softmax = True
     r1 = False
     r2 = False
@@ -233,13 +233,13 @@ class Qwen2_5_1_5B(LLMModelConfig):
 
     num_sharding = 1
     # quant config
-    ptq = QuantDtype.use_16a8w
-    group_size = None
+    ptq = QuantDtype.use_16a4w_block
+    group_size = 16
     masked_softmax = True
     r1 = False
     r2 = False
     r3 = True
-    custom_annotation = ()
+    custom_annotation = (annotate_output_16a8w,)
 
 
 @register_llm_model("qwen3-0_6b")