[BugFix][Fusion] Fix graph fusion failure problem (#5253)

Currently, the vllm pull request
(https://github.com/vllm-project/vllm/pull/24252) is causing operator
fusion to fail. This issue was previously fixed by patching the backend.
The root cause has been identified, and the problem can be resolved with
this pull request.

- vLLM version: release/v0.13.0
- vLLM main:
ad32e3e19c

---------

Signed-off-by: wxsIcey <1790571317@qq.com>
This commit is contained in:
Icey
2026-01-05 17:49:09 +08:00
committed by GitHub
parent 4a3663327b
commit e7b623b363
9 changed files with 36 additions and 267 deletions

View File

@@ -1020,8 +1020,6 @@ class TestNPUWorker(TestBase):
# Verify _dummy_run call count and order (by size descending)
expected_calls = [
unittest.mock.call(16),
unittest.mock.call(8),
unittest.mock.call(4),
unittest.mock.call(1),
]
worker.model_runner._dummy_run.assert_has_calls(expected_calls)
@@ -1030,7 +1028,7 @@ class TestNPUWorker(TestBase):
worker.model_runner.capture_model.assert_not_called()
# Verify log output
self.assertEqual(mock_logger.info.call_count, 4)
self.assertEqual(mock_logger.info.call_count, 2)
# Verify seed setting
mock_seed_everything.assert_called_once_with(12345)