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

[long_seq] remove long_seq env (#4660)

Browse Source 
### What this PR does / why we need it?
remove env VLLM_ASCEND_ENABLE_CONTEXT_PARALLEL 

- vLLM version: v0.12.0

---------

Signed-off-by: LookAround <lixushi@huawei.com>
Signed-off-by: ZhangMingWei716 <2894054457@qq.com>
Co-authored-by: ZhangMingWei716 <2894054457@qq.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
LookAround0301
2025-12-05 10:31:49 +08:00
committed by GitHub
parent ea54388e19
commit b32ef53b3b
16 changed files with 230 additions and 176 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
tests/ut/attention/test_attention_v1.py
View File

@@ -54,12 +54,16 @@ class TestAscendAttentionBackend(TestBase):
class TestAscendAttentionMetadataBuilder(TestBase):
@patch('vllm.distributed.parallel_state.get_pcp_group')
@patch('vllm.distributed.parallel_state._PCP',
new_callable=lambda: MagicMock(spec=GroupCoordinator))
@patch('vllm.distributed.parallel_state.get_dcp_group')
@patch('vllm.distributed.parallel_state._DCP',
new_callable=lambda: MagicMock(spec=GroupCoordinator))
@patch("vllm.distributed.get_decode_context_model_parallel_world_size",
return_value=1)
def setUp(self, mock_get_dcp_size, mock_dcp, mock_get_dcp_group):
def setUp(self, mock_get_dcp_size, mock_dcp, mock_get_dcp_group, mock_pcp,
mock_get_pcp_group):
mock_dcp.world_size = 1
dcp_group = MagicMock(spec=GroupCoordinator)
dcp_group.rank_in_group = 0
@@ -67,6 +71,13 @@ class TestAscendAttentionMetadataBuilder(TestBase):
dcp_group.device_group = MagicMock()
mock_get_dcp_group.return_value = dcp_group
mock_pcp.world_size = 1
pcp_group = MagicMock(spec=GroupCoordinator)
pcp_group.rank_in_group = 0
pcp_group.world_size = 1
pcp_group.device_group = MagicMock()
mock_get_pcp_group.return_value = pcp_group
self.mock_vllm_config = MagicMock()
self.mock_vllm_config.speculative_config = None
self.mock_vllm_config.model_config.max_model_len = 640
@@ -117,12 +128,16 @@ class TestAscendAttentionMetadataBuilder(TestBase):
class TestAscendAttentionBackendImpl(TestBase):
@patch('vllm.distributed.parallel_state.get_pcp_group')
@patch('vllm.distributed.parallel_state._PCP',
new_callable=lambda: MagicMock(spec=GroupCoordinator))
@patch('vllm.distributed.parallel_state.get_dcp_group')
@patch('vllm.distributed.parallel_state._DCP',
new_callable=lambda: MagicMock(spec=GroupCoordinator))
@patch("vllm.distributed.get_decode_context_model_parallel_world_size",
return_value=1)
def setUp(self, mock_get_dcp_size, mock_dcp, mock_get_dcp_group):
def setUp(self, mock_get_dcp_size, mock_dcp, mock_get_dcp_group, mock_pcp,
mock_get_pcp_group):
mock_dcp.world_size = 1
dcp_group = MagicMock(spec=GroupCoordinator)
dcp_group.rank_in_group = 0
@@ -130,6 +145,13 @@ class TestAscendAttentionBackendImpl(TestBase):
dcp_group.device_group = MagicMock()
mock_get_dcp_group.return_value = dcp_group
mock_pcp.world_size = 1
pcp_group = MagicMock(spec=GroupCoordinator)
pcp_group.rank_in_group = 0
pcp_group.world_size = 1
pcp_group.device_group = MagicMock()
mock_get_pcp_group.return_value = pcp_group
self.layer = MagicMock()
self.layer.layer_name = "test_layer"
self.layer._k_scale_float = 1.0