From 9d6f0372e50ac29b1adda94d29646069acd85597 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Wed, 20 Aug 2025 13:31:41 +0000
Subject: [PATCH] Fix duplicate torch operation registrations in tests/compile

Co-authored-by: ProExpertProg <11367180+ProExpertProg@users.noreply.github.com>
---
 tests/compile/piecewise/test_multiple_graphs.py | 4 ++--
 tests/compile/piecewise/test_simple.py          | 6 +++---
 tests/compile/piecewise/test_toy_llama.py       | 4 ++--
 3 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/tests/compile/piecewise/test_multiple_graphs.py b/tests/compile/piecewise/test_multiple_graphs.py
index e460d70951786..b71ae66d3fd28 100644
--- a/tests/compile/piecewise/test_multiple_graphs.py
+++ b/tests/compile/piecewise/test_multiple_graphs.py
@@ -19,7 +19,7 @@ from vllm.forward_context import set_forward_context
 from vllm.utils import direct_register_custom_op
 
 # create a library to hold the custom op
-silly_lib = Library("silly", "FRAGMENT")  # noqa
+silly_lib = Library("silly_multiple", "FRAGMENT")  # noqa
 
 BATCH_SIZE = 32
 MLP_SIZE = 128
@@ -188,7 +188,7 @@ def test_ignore_torch_compile_decorator():
         def forward(self, x: torch.Tensor) -> torch.Tensor:
             x = x + x
             attn_output = torch.empty_like(x)
-            torch.ops.silly.attention(x, x, x, attn_output)
+            torch.ops.silly_multiple.attention(x, x, x, attn_output)
             x = attn_output
             x = x * 3
             return x
diff --git a/tests/compile/piecewise/test_simple.py b/tests/compile/piecewise/test_simple.py
index 2d1a72d44ec70..2920d2cdd7ae7 100644
--- a/tests/compile/piecewise/test_simple.py
+++ b/tests/compile/piecewise/test_simple.py
@@ -20,7 +20,7 @@ from vllm.utils import direct_register_custom_op
 global_counter = 0
 
 # create a library to hold the custom op
-silly_lib = Library("silly", "FRAGMENT")  # noqa
+silly_lib = Library("silly_simple", "FRAGMENT")  # noqa
 
 
 def silly_attention(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor,
@@ -66,12 +66,12 @@ class SillyModel(nn.Module):
         x = x + 1
         x = x + 2
         out = torch.empty_like(x)
-        torch.ops.silly.attention(x, x, x, out)
+        torch.ops.silly_simple.attention(x, x, x, out)
         x = out
         x = x - 2
         x = x - 1
         out = torch.empty_like(x)
-        torch.ops.silly.attention(x, x, x, out)
+        torch.ops.silly_simple.attention(x, x, x, out)
         x = out
         x = x + 1
         return x
diff --git a/tests/compile/piecewise/test_toy_llama.py b/tests/compile/piecewise/test_toy_llama.py
index bcfd0d834c5db..0e7ab819c65e8 100644
--- a/tests/compile/piecewise/test_toy_llama.py
+++ b/tests/compile/piecewise/test_toy_llama.py
@@ -24,7 +24,7 @@ from vllm.forward_context import BatchDescriptor, set_forward_context
 from vllm.utils import direct_register_custom_op
 
 # create a library to hold the custom op
-silly_lib = Library("silly", "FRAGMENT")  # noqa
+silly_lib = Library("silly_toy_llama", "FRAGMENT")  # noqa
 
 
 def silly_attention(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor,
@@ -160,7 +160,7 @@ class LlamaAttention(nn.Module):
         k = k + positions.unsqueeze(1)
 
         attn_output = torch.empty_like(q)
-        torch.ops.silly.attention(q, k, v, attn_output)
+        torch.ops.silly_toy_llama.attention(q, k, v, attn_output)
 
         output = self.output_projection(attn_output)
         return output