EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[CI] Fix FusedMoEConfig and input batch failure to recover CI (#1602)

Browse Source 
Make CI happy

1.
c1909e7e8c
changed moeConfig init way
2.
48fb076cbc
changed input batch logic.

This PR address these change to vllm-ascend.

Closes: https://github.com/vllm-project/vllm-ascend/issues/1600

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-07-03 18:36:17 +08:00
committed by GitHub
parent d96da1f00c
commit a45dfde283
11 changed files with 173 additions and 134 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/ut/patch/worker/patch_common/test_patch_sampler.py
View File

@@ -12,8 +12,8 @@ class TestTopKTopPSamplerOptimize(unittest.TestCase):
@mock.patch.dict(os.environ, {"VLLM_ASCEND_ENABLE_TOPK_OPTIMIZE": "1"})
@mock.patch("torch_npu.npu_top_k_top_p")
def test_npu_topk_topp_called_when_optimized(self, mock_npu_op):
import vllm_ascend.patch.worker.patch_common.patch_sampler
importlib.reload(vllm_ascend.patch.worker.patch_common.patch_sampler)
import vllm_ascend.patch.worker.patch_0_9_1.patch_sampler
importlib.reload(vllm_ascend.patch.worker.patch_0_9_1.patch_sampler)
mock_npu_op.return_value = (torch.randn(1, 3))
sampler = topk_topp_sampler.TopKTopPSampler()