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

Drop torchair (#4814)

Browse Source 
aclgraph is stable and fast now. Let's drop torchair graph mode now.

TODO: some logic to adapt torchair should be cleaned up as well. We'll
do it in the following PR.

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Co-authored-by: Mengqing Cao <cmq0113@163.com>
This commit is contained in:
wangxiyuan
2025-12-10 09:20:40 +08:00
committed by GitHub
parent ba9cda9dfd
commit 835b4c8f1d
84 changed files with 77 additions and 16881 deletions
Download Patch File Download Diff File Expand all files Collapse all files

112
tests/ut/test_platform.py
View File

@@ -31,7 +31,6 @@ class TestNPUPlatform(TestBase):
@staticmethod
def mock_vllm_ascend_config():
mock_ascend_config = MagicMock()
mock_ascend_config.torchair_graph_config.enabled = False
mock_ascend_config.xlite_graph_config.enabled = False
mock_ascend_config.enable_shared_expert_dp = False
return mock_ascend_config
@@ -403,47 +402,6 @@ class TestNPUPlatform(TestBase):
CUDAGraphMode.NONE,
)
@patch('vllm_ascend.utils.get_ascend_device_type',
return_value=AscendDeviceType._910_93)
@patch("vllm_ascend.utils.update_default_aclgraph_sizes")
@patch("vllm_ascend.ascend_config.check_ascend_config")
@patch("vllm_ascend.ascend_config.init_ascend_config")
@patch(
"vllm_ascend.core.recompute_schedule_config.RecomputeSchedulerConfig.initialize_from_config"
)
def test_check_and_update_config_torchair_enabled_compilation(
self, mock_init_recompute, mock_init_ascend, mock_check_ascend,
mock_update_default, mock_soc_version):
mock_update_default.return_value = MagicMock()
mock_ascend_config = TestNPUPlatform.mock_vllm_ascend_config()
mock_ascend_config.torchair_graph_config.enabled = True
mock_init_ascend.return_value = mock_ascend_config
vllm_config = TestNPUPlatform.mock_vllm_config()
vllm_config.model_config.enforce_eager = False
vllm_config.parallel_config.decode_context_parallel_size = 1
vllm_config.parallel_config.prefill_context_parallel_size = 1
vllm_config.parallel_config.tensor_parallel_size = 1
mock_init_recompute.return_value = MagicMock()
vllm_config.scheduler_config = MagicMock()
vllm_config.compilation_config.mode = CompilationMode.VLLM_COMPILE
with self.assertLogs(logger="vllm", level="INFO") as cm:
from vllm_ascend import platform
importlib.reload(platform)
self.platform.check_and_update_config(vllm_config)
self.assertTrue("Torchair compilation enabled" in cm.output[0])
self.assertEqual(
vllm_config.compilation_config.mode,
CompilationMode.NONE,
)
self.assertEqual(
vllm_config.compilation_config.cudagraph_mode,
CUDAGraphMode.NONE,
)
@patch('vllm_ascend.utils.get_ascend_device_type',
return_value=AscendDeviceType._910_93)
@patch("vllm_ascend.ascend_config.check_ascend_config")
@@ -503,16 +461,6 @@ class TestNPUPlatform(TestBase):
"vllm_ascend.worker.worker_v1.NPUWorker",
)
test_ascend_config = TestNPUPlatform.mock_vllm_ascend_config()
test_ascend_config.torchair_graph_config.enabled = True
mock_init_ascend.return_value = test_ascend_config
vllm_config.parallel_config.worker_cls = "auto"
self.platform.check_and_update_config(vllm_config)
self.assertEqual(
vllm_config.parallel_config.worker_cls,
"vllm_ascend.torchair.torchair_worker.NPUTorchairWorker",
)
test_ascend_config = TestNPUPlatform.mock_vllm_ascend_config()
test_ascend_config.xlite_graph_config.enabled = True
mock_init_ascend.return_value = test_ascend_config
@@ -550,14 +498,7 @@ class TestNPUPlatform(TestBase):
self.platform.check_and_update_config(vllm_config)
self.assertEqual(vllm_config.compilation_config.custom_ops, [])
@patch('vllm_ascend.platform.get_ascend_config')
def test_get_attn_backend_cls_use_v1_and_mla(self, mock_get_ascend_config):
mock_config = MagicMock()
mock_config.torchair_graph_config.enabled = False
mock_config.enable_shared_expert_dp = False
mock_get_ascend_config.return_value = mock_config
def test_get_attn_backend_cls_use_v1_and_mla(self):
result = self.platform.get_attn_backend_cls(
selected_backend="ascend",
head_size=64,
@@ -570,56 +511,7 @@ class TestNPUPlatform(TestBase):
self.assertEqual(result,
"vllm_ascend.attention.mla_v1.AscendMLABackend")
@patch('vllm_ascend.platform.get_ascend_config')
def test_get_attn_backend_cls_use_v1_mla_and_torchair(
self, mock_get_ascend_config):
mock_config = MagicMock()
mock_config.torchair_graph_config.enabled = True
mock_get_ascend_config.return_value = mock_config
result = self.platform.get_attn_backend_cls(
selected_backend="ascend",
head_size=64,
dtype="float16",
kv_cache_dtype="float16",
block_size=64,
#use_sfa=False,
use_mla=True,
)
self.assertEqual(
result,
"vllm_ascend.torchair.torchair_mla.AscendMLATorchairBackend")
@patch('vllm_ascend.platform.get_ascend_config')
def test_get_attn_backend_cls_use_v1_and_torchair(self,
mock_get_ascend_config):
mock_config = MagicMock()
mock_config.torchair_graph_config.enabled = True
mock_get_ascend_config.return_value = mock_config
result = self.platform.get_attn_backend_cls(
selected_backend="ascend",
head_size=64,
dtype="float16",
kv_cache_dtype="float16",
block_size=64,
#use_sfa=False,
use_mla=False,
)
self.assertEqual(
result,
"vllm_ascend.torchair.torchair_attention.AscendAttentionTorchairBackend"
)
@patch('vllm_ascend.platform.get_ascend_config')
def test_get_attn_backend_cls_use_v1_only(self, mock_get_ascend_config):
mock_config = MagicMock()
mock_config.torchair_graph_config.enabled = False
mock_get_ascend_config.return_value = mock_config
def test_get_attn_backend_cls_use_v1_only(self):
result = self.platform.get_attn_backend_cls(
selected_backend="ascend",
head_size=64,