EngineX-Ascend/enginex-ascend-910-vllm
10
0
Fork 1
Code Issues Pull Requests Actions Projects Releases Wiki Activity

init v0.11.0rc0

Browse Source
This commit is contained in:
luopingyi
2025-10-14 10:38:28 +08:00
parent 67afd0ea78
commit 66dc16f966
278 changed files with 28130 additions and 11708 deletions
Download Patch File Download Diff File Expand all files Collapse all files

59
tests/ut/test_ascend_config.py
View File

@@ -43,6 +43,7 @@ class TestAscendConfig(TestBase):
# No additional config given, check the default value here.
ascend_config = init_ascend_config(test_vllm_config)
self.assertIsNone(ascend_config.expert_map_path)
self.assertFalse(ascend_config.multistream_overlap_shared_expert)
torchair_graph_config = ascend_config.torchair_graph_config
self.assertFalse(torchair_graph_config.enabled)
@@ -51,8 +52,8 @@ class TestAscendConfig(TestBase):
self.assertEqual(torchair_graph_config.graph_batch_sizes, [])
self.assertFalse(torchair_graph_config.graph_batch_sizes_init)
self.assertFalse(torchair_graph_config.enable_multistream_mla)
self.assertFalse(torchair_graph_config.enable_multistream_moe)
self.assertTrue(torchair_graph_config.enable_view_optimize)
self.assertTrue(torchair_graph_config.enable_frozen_parameter)
self.assertFalse(torchair_graph_config.enable_kv_nz)
ascend_scheduler_config = ascend_config.ascend_scheduler_config
@@ -68,10 +69,11 @@ class TestAscendConfig(TestBase):
"graph_batch_sizes": [1, 2, 4],
"graph_batch_sizes_init": False,
"enable_multistream_mla": True,
"enable_multistream_moe": True,
"enable_view_optimize": True,
"enable_frozen_parameter": True,
"enable_kv_nz": True
},
"multistream_overlap_shared_expert": True,
"ascend_scheduler_config": {
"enabled": True
},
@@ -80,6 +82,7 @@ class TestAscendConfig(TestBase):
}
ascend_config = init_ascend_config(test_vllm_config)
self.assertEqual(ascend_config.expert_map_path, "test_expert_map_path")
self.assertTrue(ascend_config.multistream_overlap_shared_expert)
torchair_graph_config = ascend_config.torchair_graph_config
self.assertTrue(torchair_graph_config.enabled)
@@ -87,8 +90,8 @@ class TestAscendConfig(TestBase):
self.assertEqual(torchair_graph_config.graph_batch_sizes, [1, 2, 4])
self.assertFalse(torchair_graph_config.graph_batch_sizes_init)
self.assertTrue(torchair_graph_config.enable_multistream_mla)
self.assertTrue(torchair_graph_config.enable_multistream_moe)
self.assertTrue(torchair_graph_config.enable_view_optimize)
self.assertTrue(torchair_graph_config.enable_frozen_parameter)
self.assertTrue(torchair_graph_config.enable_kv_nz)
ascend_scheduler_config = ascend_config.ascend_scheduler_config
@@ -215,21 +218,6 @@ class TestAscendConfig(TestBase):
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),
@@ -318,17 +306,6 @@ class TestAscendConfig(TestBase):
}
init_ascend_config(test_vllm_config)
# enable_multistream_moe should not be enabled without torchair graph mode
with self.assertRaises(RuntimeError):
test_vllm_config.additional_config = {
"torchair_graph_config": {
"enabled": False,
"enable_multistream_moe": True,
},
"refresh": True
}
init_ascend_config(test_vllm_config)
# mode should not be configured without torchair graph mode
with self.assertRaises(RuntimeError):
test_vllm_config.additional_config = {
@@ -359,3 +336,27 @@ class TestAscendConfig(TestBase):
test_vllm_config.parallel_config = ParallelConfig(
data_parallel_size=4, tensor_parallel_size=2)
init_ascend_config(test_vllm_config)
with self.assertRaises(AssertionError):
test_vllm_config.additional_config = {
"torchair_graph_config": {
"enabled": True,
},
"oproj_tensor_parallel_size": 2,
"refresh": True
}
test_vllm_config.parallel_config = ParallelConfig(
data_parallel_size=4, tensor_parallel_size=2)
init_ascend_config(test_vllm_config)
with self.assertRaises(AssertionError):
test_vllm_config.additional_config = {
"torchair_graph_config": {
"enabled": False,
},
"oproj_tensor_parallel_size": 2,
"refresh": True
}
test_vllm_config.parallel_config = ParallelConfig(
data_parallel_size=4, tensor_parallel_size=1)
init_ascend_config(test_vllm_config)