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

[refactor](UT,PCP,DCP) refactor pcp&dcp patches in UTs (#5505)

Browse Source 
### What this PR does / why we need it?
Refactor PCP & DCP patches in UTs: Merge and reuse communication groups
and communication function patches to reduce code duplication.
### Does this PR introduce _any_ user-facing change?
No
- vLLM version: v0.13.0
- vLLM main:
45c1ca1ca1

Signed-off-by: QiuChunshuo <qiuchunshuo@huawei.com>
This commit is contained in:
Qiu
2026-01-05 09:05:45 +08:00
committed by GitHub
parent 46c2fc6a3c
commit 96775a27a8
6 changed files with 128 additions and 531 deletions
Download Patch File Download Diff File Expand all files Collapse all files

51
tests/ut/attention/test_attention_v1.py
View File

@@ -1,7 +1,6 @@
from unittest.mock import MagicMock, patch
import torch
from vllm.distributed.parallel_state import GroupCoordinator
from tests.ut.base import TestBase
from vllm_ascend.attention.attention_v1 import (AscendAttentionBackend,
@@ -50,30 +49,7 @@ 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, mock_pcp,
mock_get_pcp_group):
mock_dcp.world_size = 1
dcp_group = MagicMock(spec=GroupCoordinator)
dcp_group.rank_in_group = 0
dcp_group.world_size = 1
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
def setUp(self):
self.mock_vllm_config = MagicMock()
self.mock_vllm_config.speculative_config = None
self.mock_vllm_config.model_config.max_model_len = 640
@@ -126,30 +102,7 @@ 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, mock_pcp,
mock_get_pcp_group):
mock_dcp.world_size = 1
dcp_group = MagicMock(spec=GroupCoordinator)
dcp_group.rank_in_group = 0
dcp_group.world_size = 1
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
def setUp(self):
self.layer = MagicMock()
self.layer.layer_name = "test_layer"
self.layer._k_scale_float = 1.0