xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-11 06:45:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[ROCm][CI] Fix test_cudagraph_mode failure in AMD CI (#29367)

Browse Source 
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
This commit is contained in:
Micah Williamson 2025-11-25 01:55:09 -06:00 committed by GitHub
parent 12c007e288
commit ef1f7030f0
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 51 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
tests/v1/attention/utils.py
View File

@ -340,4 +340,11 @@ full_cg_backend_configs = {
"cudagraph_mode": "FULL_AND_PIECEWISE", "cudagraph_mode": "FULL_AND_PIECEWISE",
}, },
), ),
"RocmAttn": BackendConfig(
name="RocmAttn",
env_vars={"VLLM_V1_USE_PREFILL_DECODE_ATTENTION": "1"},
comp_config={
"cudagraph_mode": "FULL",
},
),
} }

62
tests/v1/cudagraph/test_cudagraph_mode.py
View File

@ -35,14 +35,22 @@ def temporary_environ(env_vars):
# test attention backend and cudagraph_mode combo # test attention backend and cudagraph_mode combo
# (backend_name, cudagraph_mode, supported) # (backend_name, cudagraph_mode, supported)
combo_cases_1 = [ if current_platform.is_rocm():
("FA3", "FULL", True), combo_cases_1 = [
("FA3", "FULL_AND_PIECEWISE", True), ("RocmAttn", "FULL", True),
("FA2", "FULL", True), # Should fallback to FULL_AND_PIECEWISE ("RocmAttn", "FULL_AND_PIECEWISE", True),
("FA2", "FULL_AND_PIECEWISE", True), ("TritonAttn", "FULL", True),
("FlashInfer", "FULL", True), # Should fallback to FULL_AND_PIECEWISE ("TritonAttn", "FULL_AND_PIECEWISE", True),
("FlashInfer", "FULL_AND_PIECEWISE", True), ]
] else:
combo_cases_1 = [
("FA3", "FULL", True),
("FA3", "FULL_AND_PIECEWISE", True),
("FA2", "FULL", True), # Should fallback to FULL_AND_PIECEWISE
("FA2", "FULL_AND_PIECEWISE", True),
("FlashInfer", "FULL", True), # Should fallback to FULL_AND_PIECEWISE
("FlashInfer", "FULL_AND_PIECEWISE", True),
]
@pytest.mark.parametrize("backend_name, cudagraph_mode, supported", combo_cases_1) @pytest.mark.parametrize("backend_name, cudagraph_mode, supported", combo_cases_1)
@ -92,18 +100,32 @@ def test_backend_and_cudagraph_mode_combo(backend_name, cudagraph_mode, supporte
# test cudagraph_mode with different compilation mode. # test cudagraph_mode with different compilation mode.
# (backend_name, cudagraph_mode, compilation_mode, supported) # (backend_name, cudagraph_mode, compilation_mode, supported)
combo_cases_2 = [ if current_platform.is_rocm():
("FA2", "FULL", CompilationMode.NONE, True), combo_cases_2 = [
("FA2", "FULL", CompilationMode.VLLM_COMPILE, True), ("RocmAttn", "FULL", CompilationMode.NONE, True),
("FA2", "PIECEWISE", CompilationMode.NONE, False), ("RocmAttn", "FULL", CompilationMode.VLLM_COMPILE, True),
("FA2", "PIECEWISE", CompilationMode.VLLM_COMPILE, True), ("RocmAttn", "PIECEWISE", CompilationMode.NONE, False),
("FA2", "FULL_AND_PIECEWISE", CompilationMode.NONE, False), ("RocmAttn", "PIECEWISE", CompilationMode.VLLM_COMPILE, True),
("FA2", "FULL_AND_PIECEWISE", CompilationMode.VLLM_COMPILE, True), ("RocmAttn", "FULL_AND_PIECEWISE", CompilationMode.NONE, False),
("FA2", "FULL_DECODE_ONLY", CompilationMode.NONE, True), ("RocmAttn", "FULL_AND_PIECEWISE", CompilationMode.VLLM_COMPILE, True),
("FA2", "FULL_DECODE_ONLY", CompilationMode.VLLM_COMPILE, True), ("RocmAttn", "FULL_DECODE_ONLY", CompilationMode.NONE, True),
("FA2", "NONE", CompilationMode.NONE, True), ("RocmAttn", "FULL_DECODE_ONLY", CompilationMode.VLLM_COMPILE, True),
("FA2", "NONE", CompilationMode.VLLM_COMPILE, True), ("RocmAttn", "NONE", CompilationMode.NONE, True),
] ("RocmAttn", "NONE", CompilationMode.VLLM_COMPILE, True),
]
else:
combo_cases_2 = [
("FA2", "FULL", CompilationMode.NONE, True),
("FA2", "FULL", CompilationMode.VLLM_COMPILE, True),
("FA2", "PIECEWISE", CompilationMode.NONE, False),
("FA2", "PIECEWISE", CompilationMode.VLLM_COMPILE, True),
("FA2", "FULL_AND_PIECEWISE", CompilationMode.NONE, False),
("FA2", "FULL_AND_PIECEWISE", CompilationMode.VLLM_COMPILE, True),
("FA2", "FULL_DECODE_ONLY", CompilationMode.NONE, True),
("FA2", "FULL_DECODE_ONLY", CompilationMode.VLLM_COMPILE, True),
("FA2", "NONE", CompilationMode.NONE, True),
("FA2", "NONE", CompilationMode.VLLM_COMPILE, True),
]
@pytest.mark.parametrize( @pytest.mark.parametrize(

4
vllm/platforms/rocm.py
View File

@ -321,8 +321,8 @@ class RocmPlatform(Platform):
return AttentionBackendEnum.TRITON_ATTN.get_path() return AttentionBackendEnum.TRITON_ATTN.get_path()
raise RuntimeError( raise RuntimeError(
"V0 attention backends have been removed. Set VLLM_USE_V1=1 " f"Attention backend {selected_backend.name} is not supported on "
"to select a supported backend." "ROCm. Note that V0 attention backends have been removed."
) )
@classmethod @classmethod