EngineX-Ascend/enginex-ascend-910-vllm
11
0
Fork 1
Code Issues Pull Requests Actions Projects Releases Wiki Activity

init v0.23.0

Browse Source 
Signed-off-by: Sun Ruoxi <sunruoxi@4paradigm.com>
This commit is contained in:
Sun Ruoxi
2026-08-27 15:11:51 +08:00
parent b582a8e7d1
commit 7f8a1b1f7a
2849 changed files with 712887 additions and 22001 deletions
Download Patch File Download Diff File Expand all files Collapse all files

49
tests/ut/distributed/test_communicator.py
View File

@@ -4,19 +4,14 @@ from unittest.mock import MagicMock, patch
import torch
import torch.distributed as dist
from vllm_ascend.distributed.communicator import NPUCommunicator
from vllm_ascend.distributed.device_communicators.npu_communicator import NPUCommunicator
class TestNPUCommunicator(unittest.TestCase):
@patch("vllm.config.get_current_vllm_config", return_value=None)
@patch("torch.npu.current_device", return_value=MagicMock())
@patch("torch.npu.set_device", return_value=MagicMock())
@patch("torch.distributed.get_process_group_ranks",
return_value={
0: 0,
1: 1
})
@patch("torch.distributed.get_process_group_ranks", return_value={0: 0, 1: 1})
@patch("torch.distributed.get_group_rank", return_value={0: 0, 1: 1})
@patch("torch.distributed.is_initialized", return_value=True)
@patch("torch.distributed.get_rank", return_value=1)
@@ -27,15 +22,8 @@ class TestNPUCommunicator(unittest.TestCase):
@patch("torch.distributed.get_process_group_ranks", return_value=[0, 1])
@patch("torch.npu.device")
def test_all_to_all_with_sizes(self, *_):
def patched_all_to_all(output_tensor_list,
input_tensor_list,
group=None,
async_op=False):
output_tensor_list[:] = ([
torch.tensor([10, 20]),
torch.tensor([50, 60])
])
def patched_all_to_all(output_tensor_list, input_tensor_list, group=None, async_op=False):
output_tensor_list[:] = [torch.tensor([10, 20]), torch.tensor([50, 60])]
torch.distributed.all_to_all = patched_all_to_all
@@ -43,22 +31,17 @@ class TestNPUCommunicator(unittest.TestCase):
gather_sizes = [2, 2]
input_ = torch.tensor([10, 20, 30, 40])
comm = NPUCommunicator(cpu_group=dist.group.WORLD)
with patch.dict(dist.distributed_c10d._world.pg_map, {dist.group.WORLD: MagicMock()}, clear=False):
comm = NPUCommunicator(cpu_group=dist.group.WORLD)
output = comm.all_to_all(input_,
scatter_sizes=scatter_sizes,
gather_sizes=gather_sizes)
output = comm.all_to_all(input_, scatter_sizes=scatter_sizes, gather_sizes=gather_sizes)
assert output.tolist() == [10, 20, 50, 60]
@patch("vllm.config.get_current_vllm_config", return_value=None)
@patch("torch.npu.current_device", return_value=MagicMock())
@patch("torch.npu.set_device", return_value=MagicMock())
@patch("torch.distributed.get_process_group_ranks",
return_value={
0: 0,
1: 1
})
@patch("torch.distributed.get_process_group_ranks", return_value={0: 0, 1: 1})
@patch("torch.distributed.get_group_rank", return_value={0: 0, 1: 1})
@patch("torch.distributed.is_initialized", return_value=True)
@patch("torch.distributed.get_rank", return_value=1)
@@ -69,21 +52,15 @@ class TestNPUCommunicator(unittest.TestCase):
@patch("torch.distributed.get_process_group_ranks", return_value=[0, 1])
@patch("torch.npu.device")
def test_all_to_all_without_sizes(self, *_):
def patched_all_to_all(output_tensor_list,
input_tensor_list,
group=None,
async_op=False):
output_tensor_list[:] = ([
torch.tensor([[10, 20]]),
torch.tensor([[50, 60]])
])
def patched_all_to_all(output_tensor_list, input_tensor_list, group=None, async_op=False):
output_tensor_list[:] = [torch.tensor([[10, 20]]), torch.tensor([[50, 60]])]
torch.distributed.all_to_all = patched_all_to_all
input_ = torch.tensor([[10, 20], [30, 40]])
comm = NPUCommunicator(cpu_group=dist.group.WORLD)
output = comm.all_to_all(input_, scatter_dim=0, gather_dim=0)
with patch.dict(dist.distributed_c10d._world.pg_map, {dist.group.WORLD: MagicMock()}, clear=False):
comm = NPUCommunicator(cpu_group=dist.group.WORLD)
output = comm.all_to_all(input_, scatter_dim=0, gather_dim=0)
assert output.tolist() == [[10, 20], [50, 60]]