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

[Misc] Remove VLLM_USE_V1 usage in code (#1764)

Browse Source 
We plan to remove V0 code from this version. The first step is to delete
v0 usage.

Related: https://github.com/vllm-project/vllm-ascend/issues/1620

- vLLM version: v0.9.2
- vLLM main:
61e20828da

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-07-15 11:52:16 +08:00
committed by GitHub
parent 494b0f474f
commit 7bdada58eb
6 changed files with 100 additions and 217 deletions
Download Patch File Download Diff File Expand all files Collapse all files

107
tests/ut/test_ascend_config.py
View File

@@ -193,71 +193,48 @@ class TestAscendConfig(TestBase):
@_clean_up_ascend_config
def test_check_ascend_config_wrong_case(self):
test_vllm_config = VllmConfig()
# For V0 engine
with mock.patch.dict(os.environ, {"VLLM_USE_V1": "0"}):
with self.assertRaises(NotImplementedError):
test_vllm_config.additional_config = {
"torchair_graph_config": {
"enabled": True,
},
"refresh": True
}
init_ascend_config(test_vllm_config)
check_ascend_config(test_vllm_config, False)
with self.assertRaises(NotImplementedError):
test_vllm_config.additional_config = {
"ascend_scheduler_config": {
"enabled": True,
},
"refresh": True
}
init_ascend_config(test_vllm_config)
check_ascend_config(test_vllm_config, True)
# For V1 engine
with mock.patch.dict(os.environ, {"VLLM_USE_V1": "1"}):
# torchair + eager mode
with self.assertRaises(RuntimeError):
test_vllm_config.additional_config = {
"torchair_graph_config": {
"enabled": True,
},
"refresh": True
}
init_ascend_config(test_vllm_config)
enforce_eager = True
check_ascend_config(test_vllm_config, enforce_eager)
# torchair + non deepseek model
with self.assertRaises(NotImplementedError):
test_vllm_config.additional_config = {
"torchair_graph_config": {
"enabled": True,
},
"refresh": True
}
model_path = os.path.join(os.path.dirname(__file__),
"fake_weight")
fake_model_config = ModelConfig(model=model_path)
fake_model_config.hf_config = PretrainedConfig()
fake_model_config.hf_config.model_type = "llama"
test_vllm_config.model_config = fake_model_config
init_ascend_config(test_vllm_config)
check_ascend_config(test_vllm_config, False)
# aclgraph + deepseek model
with self.assertRaises(NotImplementedError):
test_vllm_config.additional_config = {
"torchair_graph_config": {
"enabled": False,
},
"refresh": True
}
model_path = os.path.join(os.path.dirname(__file__),
"fake_weight")
fake_model_config = ModelConfig(model=model_path)
fake_model_config.hf_config = PretrainedConfig()
fake_model_config.hf_config.model_type = "deepseek"
test_vllm_config.model_config = fake_model_config
init_ascend_config(test_vllm_config)
check_ascend_config(test_vllm_config, False)
# torchair + eager mode
with self.assertRaises(RuntimeError):
test_vllm_config.additional_config = {
"torchair_graph_config": {
"enabled": True,
},
"refresh": True
}
init_ascend_config(test_vllm_config)
enforce_eager = True
check_ascend_config(test_vllm_config, enforce_eager)
# torchair + non deepseek model
with self.assertRaises(NotImplementedError):
test_vllm_config.additional_config = {
"torchair_graph_config": {
"enabled": True,
},
"refresh": True
}
model_path = os.path.join(os.path.dirname(__file__), "fake_weight")
fake_model_config = ModelConfig(model=model_path)
fake_model_config.hf_config = PretrainedConfig()
fake_model_config.hf_config.model_type = "llama"
test_vllm_config.model_config = fake_model_config
init_ascend_config(test_vllm_config)
check_ascend_config(test_vllm_config, False)
# aclgraph + deepseek model
with self.assertRaises(NotImplementedError):
test_vllm_config.additional_config = {
"torchair_graph_config": {
"enabled": False,
},
"refresh": True
}
model_path = os.path.join(os.path.dirname(__file__), "fake_weight")
fake_model_config = ModelConfig(model=model_path)
fake_model_config.hf_config = PretrainedConfig()
fake_model_config.hf_config.model_type = "deepseek"
test_vllm_config.model_config = fake_model_config
init_ascend_config(test_vllm_config)
check_ascend_config(test_vllm_config, False)
def test_check_torchair_supported(self):
test_cases = [('deepseek_v3', True), ('PanguProMoE', True),

63
tests/ut/test_platform.py
View File

@@ -389,69 +389,6 @@ class TestNPUPlatform(TestBase):
"vllm_ascend.worker.worker_v1.NPUWorker",
)
@patch("vllm_ascend.ascend_config.check_ascend_config")
@patch("vllm_ascend.ascend_config.init_ascend_config")
@patch("vllm.envs.VLLM_USE_V1", False)
def test_check_and_update_config_speculative_worker_config(
self, mock_init_ascend, mock_check_ascend):
mock_init_ascend.return_value = self.mock_ascend_config
self.mock_vllm_config.speculative_config = MagicMock()
self.mock_vllm_config.speculative_config.disable_logprobs = True
self.mock_vllm_config.parallel_config.worker_cls = "auto"
with patch.dict("os.environ", {}):
from vllm_ascend import platform
importlib.reload(platform)
self.platform.check_and_update_config(self.mock_vllm_config)
import os
self.assertEqual(os.environ.get("ACL_OP_INIT_MODE"), "1")
self.assertEqual(
self.mock_vllm_config.parallel_config.worker_cls,
"vllm.spec_decode.spec_decode_worker.create_spec_worker",
)
self.assertEqual(
self.mock_vllm_config.parallel_config.sd_worker_cls,
"vllm_ascend.worker.worker.NPUWorker",
)
@patch("vllm_ascend.ascend_config.check_ascend_config")
@patch("vllm_ascend.ascend_config.init_ascend_config")
@patch("vllm.envs.VLLM_USE_V1", False)
def test_check_and_update_config_multi_step_worker_config(
self, mock_init_ascend, mock_check_ascend):
mock_init_ascend.return_value = self.mock_ascend_config
self.mock_vllm_config.scheduler_config.is_multi_step = True
self.mock_vllm_config.parallel_config.worker_cls = "auto"
from vllm_ascend import platform
importlib.reload(platform)
self.platform.check_and_update_config(self.mock_vllm_config)
self.assertEqual(
self.mock_vllm_config.parallel_config.worker_cls,
"vllm_ascend.worker.multi_step_worker.MultiStepWorker",
)
@patch("vllm_ascend.ascend_config.check_ascend_config")
@patch("vllm_ascend.ascend_config.init_ascend_config")
@patch("vllm.envs.VLLM_USE_V1", False)
def test_check_and_update_config_default_worker_config(
self, mock_init_ascend, mock_check_ascend):
mock_init_ascend.return_value = self.mock_ascend_config
self.mock_vllm_config.parallel_config.worker_cls = "auto"
self.mock_vllm_config.scheduler_config.is_multi_step = False
from vllm_ascend import platform
importlib.reload(platform)
self.platform.check_and_update_config(self.mock_vllm_config)
self.assertEqual(
self.mock_vllm_config.parallel_config.worker_cls,
"vllm_ascend.worker.worker.NPUWorker",
)
@patch("vllm_ascend.ascend_config.check_ascend_config")
@patch("vllm_ascend.ascend_config.init_ascend_config")
@patch("vllm_ascend.utils.is_310p", return_value=True)