【Hackathon 9th No.127】feat: Implement torch._C._nn.linear to torch.nn.functional.linear conversion by Dayuxiaoshui · Pull Request #353 · PaddlePaddle/GraphNet

Dayuxiaoshui · 2025-11-10T10:25:09Z

Add _impl_unstable_to_stable_linear_to_functional_linear method using AST transformation
Adapt to new architecture with impl_unstable_to_stable prefix

PR Category

Description

paddle-bot · 2025-11-10T10:25:19Z

Thanks for your contribution!

Dayuxiaoshui · 2025-11-10T10:31:35Z

Dayuxiaoshui · 2025-11-10T10:41:05Z

修改 log2json.py 是为了确保 JSON 报告包含完整且正确的状态和性能数据

lixinqi · 2025-11-10T10:50:43Z

graph_net/torch/backend/unstable_to_stable_backend.py

+                    if isinstance(node.func, ast.Attribute):
+                        # node.func.attr should be "linear"
+                        if node.func.attr == "linear":
+                            # node.func.value should be torch._C._nn
+                            if isinstance(node.func.value, ast.Attribute):
+                                # node.func.value.attr should be "_nn"
+                                if node.func.value.attr == "_nn":
+                                    # node.func.value.value should be torch._C
+                                    if isinstance(node.func.value.value, ast.Attribute):
+                                        # node.func.value.value.attr should be "_C"
+                                        if node.func.value.value.attr == "_C":
+                                            # node.func.value.value.value should be torch
+                                            if (
+                                                isinstance(
+                                                    node.func.value.value.value,
+                                                    ast.Name,
+                                                )
+                                                and node.func.value.value.value.id
+                                                == "torch"
+                                            ):


按如下方法改造：

filtered_nodes = [ node for node in [node] if isinstance(node.func, ast.Attribute) if node.func.attr == "linear" if isinstance(node.func.value, ast.Attribute) if node.func.value.attr == "_nn" if isinstance(node.func.value.value, ast.Attribute) if node.func.value.value.attr == "_C" if isinstance(node.func.value.value.value, ast.Name) if node.func.value.value.value.id == "torch" ] if len(filtered_nodes) > 0: ...

lixinqi · 2025-11-10T10:54:51Z

graph_net/torch/backend/unstable_to_stable_backend.py

+                gm.forward = types.MethodType(forward_func, gm)
+
+            # Update _code attribute so that gm.code returns the modified code
+            gm._code = new_code


这么做是很不正式的做法。我们单独准备了 fx_graph_serialize 机制。具体你看#352 pr 里的https://github.com/PaddlePaddle/GraphNet/pull/352/files#diff-cdf8f6acc9f3a2d0573129027aaee53e37576cdcd576205d8e0b63c740003105 这里

lixinqi · 2025-11-10T11:16:06Z

graph_net/torch/backend/unstable_to_stable_backend.py

+                gm.forward = types.MethodType(forward_func, gm)
+
+            # Update _code attribute so that gm.code returns the modified code
+            gm._code = new_code


这种直接改gm._code的做法是很不地道的，应当避免。

lixinqi · 2025-11-11T02:18:52Z

graph_net/torch/fx_graph_serialize_util.py

        # replace this line with modification code for task 123 (torch._C._nn.pad)
        # replace this line with modification code for task 125 (torch._C._nn.gelu)
        # replace this line with modification code for task 126 (torch._C._nn.scaled_dot_product_attention)
-        # replace this line with modification code for task 127 (torch._C._nn.linear)


把上面的改动放到这里，减少其他 PR 的合入冲突

lixinqi · 2025-11-11T02:19:20Z

graph_net/torch/backend/unstable_to_stable_backend.py

-        # Use serialized code to check for unstable APIs
-        graph_text = serialize_graph_module_to_str(gm)
+        # Use code to check for unstable APIs
+        graph_text = gm.code


这部分的代码不能改

check_unstable_to_stable 是裁判，裁判的逻辑不可以改。上面_impl_unstable_to_stable_linear_to_functional_linear是运动员。

lixinqi · 2025-11-11T06:52:10Z

graph_net/torch/backend/unstable_to_stable_backend.py

+            # Use serialize_graph_module_to_str to get the serialized code
+            # This ensures the code is properly serialized with unstable API replacements
+            serialized_code = serialize_graph_module_to_str(gm)
+            gm._code = serialized_code


这代码能去掉吗？

lixinqi · 2025-11-11T07:34:57Z

graph_net/torch/backend/unstable_to_stable_backend.py

把_impl_unstable_to_stable_linear_to_functional_linear函数的实现挪到这里，避开与其他 pr 的冲突

- Implement direct node.target modification for API conversion - Use serialize_graph_module_to_str for API check in check_unstable_api - Add AST-based replacement function (commented) in fx_graph_serialize_util.py - Fix log2json.py to properly initialize result field and map speedup data - Simplify conversion logic by removing complex AST code - Tested with 50 samples: 100% success rate, ES(-6) = 1.013

Dayuxiaoshui · 2025-11-11T08:19:37Z

paddle-bot bot added the contributor External developers label Nov 10, 2025

Dayuxiaoshui force-pushed the develop branch from b4c0103 to 2116c66 Compare November 10, 2025 10:30

lixinqi reviewed Nov 10, 2025

View reviewed changes

luotao1 mentioned this pull request Nov 10, 2025

【Hackathon 9th】开源贡献个人挑战赛 PaddlePaddle/Paddle#74773

Closed

lixinqi reviewed Nov 11, 2025

View reviewed changes

Dayuxiaoshui force-pushed the develop branch from d49878a to 1a77a0e Compare November 11, 2025 08:00

Dayuxiaoshui force-pushed the develop branch from 1a77a0e to 3a3b149 Compare November 11, 2025 08:01

Dayuxiaoshui and others added 2 commits November 11, 2025 16:11

Update unstable_to_stable_backend.py

fd62bcc

style: apply black formatting to unstable_to_stable_backend.py

dd55706

lixinqi approved these changes Nov 11, 2025

View reviewed changes

lixinqi merged commit 8fb4ad6 into PaddlePaddle:develop Nov 11, 2025
3 checks passed

luotao1 added the PaddlePaddle Hackathon label Nov 24, 2025

Conversation

Dayuxiaoshui commented Nov 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Category

Description

Uh oh!

paddle-bot bot commented Nov 10, 2025

Uh oh!

Dayuxiaoshui commented Nov 10, 2025

Uh oh!

Dayuxiaoshui commented Nov 10, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Dayuxiaoshui commented Nov 11, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Dayuxiaoshui commented Nov 10, 2025 •

edited

Loading