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

2
tests/ut/worker/test_input_batch.py
View File

@@ -24,8 +24,8 @@ from vllm.utils import make_tensor_with_pad
from vllm.v1.pool.metadata import PoolingMetadata
from vllm.v1.sample.logits_processor import LogitsProcessors
from vllm.v1.sample.metadata import SamplingMetadata
from vllm.v1.worker.block_table import BlockTable, MultiGroupBlockTable
from vllm_ascend.worker.block_table import BlockTable, MultiGroupBlockTable
from vllm_ascend.worker.npu_input_batch import CachedRequestState, InputBatch
VOCAB_SIZE = 1024

107
tests/ut/worker/test_model_runner_v1.py Normal file
View File

@@ -0,0 +1,107 @@
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# This file is a part of the vllm-ascend project.
from unittest.mock import MagicMock, patch
import pytest
from vllm_ascend.ascend_forward_context import MoECommType
from vllm_ascend.utils import AscendSocVersion
from vllm_ascend.worker.model_runner_v1 import NPUModelRunner
# yapf: disable
@pytest.mark.parametrize(
"soc_version, enable_expert_parallel, world_size, num_tokens, mc2_tokens_capacity, quant_type, expected_method",
[
# Case 1: Expert parallel is disabled, should always be 'allgather'
(AscendSocVersion.A2, False, 8, 100, 256, None, MoECommType.ALLGATHER),
(AscendSocVersion.A3, False, 16, 500, 256, None, MoECommType.ALLGATHER),
# Case 2: A2 SOC with w4a8_dynamic -> use alltoall when not mc2
(AscendSocVersion.A2, True, 8, 100, 256, "w4a8_dynamic", MoECommType.ALLTOALL),
(AscendSocVersion.A2, True, 16, 257, 256, "w4a8_dynamic", MoECommType.ALLTOALL),
(AscendSocVersion.A2, True, 16, 100, 256, "w4a8_dynamic", MoECommType.MC2), # meets mc2 condition
# Case 3: A2 SOC without w4a8_dynamic -> fallback to allgather
(AscendSocVersion.A2, True, 8, 100, 256, None, MoECommType.ALLGATHER),
(AscendSocVersion.A2, True, 16, 257, 256, None, MoECommType.ALLGATHER),
# Case 4: A3 SOC
(AscendSocVersion.A3, True, 8, 100, 256, None, MoECommType.MC2),
(AscendSocVersion.A3, True, 8, 257, 256, None, MoECommType.ALLTOALL),
])
# yapf: enable
def test_select_moe_comm_method(soc_version, enable_expert_parallel,
world_size, num_tokens, mc2_tokens_capacity,
quant_type, expected_method):
"""
Tests the _select_moe_comm_method with various configurations including quant_type.
"""
# Mock the NPUModelRunner instance and its dependencies
mock_runner = MagicMock(spec=NPUModelRunner)
mock_runner.parallel_config = MagicMock()
mock_runner.parallel_config.enable_expert_parallel = enable_expert_parallel
mock_runner.parallel_config.world_size_across_dp = world_size
mock_runner.mc2_tokens_capacity = mc2_tokens_capacity
# Add vllm_config.model_config.hf_config mock with moe_quantize
mock_hf_config = MagicMock()
mock_hf_config.moe_quantize = quant_type
mock_model_config = MagicMock()
mock_model_config.hf_config = mock_hf_config
mock_vllm_config = MagicMock()
mock_vllm_config.model_config = mock_model_config
mock_runner.vllm_config = mock_vllm_config
# Patch the helper functions
with patch('vllm_ascend.worker.model_runner_v1.get_ascend_soc_version',
return_value=soc_version), \
patch('vllm_ascend.worker.model_runner_v1.is_global_first_rank',
return_value=True):
# Bind the real method to the mock object
method = NPUModelRunner._select_moe_comm_method(
mock_runner, num_tokens, False)
# Assert the result
assert method == expected_method
def test_select_moe_comm_method_unsupported_soc():
"""
Tests that _select_moe_comm_method raises ValueError for an unsupported SOC.
"""
mock_runner = MagicMock(spec=NPUModelRunner)
mock_runner.parallel_config = MagicMock()
mock_runner.parallel_config.enable_expert_parallel = True
mock_runner.mc2_tokens_capacity = 256
# Add vllm_config.model_config.hf_config mock with moe_quantize
mock_hf_config = MagicMock()
mock_hf_config.moe_quantize = None
mock_model_config = MagicMock()
mock_model_config.hf_config = mock_hf_config
mock_vllm_config = MagicMock()
mock_vllm_config.model_config = mock_model_config
mock_runner.vllm_config = mock_vllm_config
unsupported_soc = "UnsupportedSOC"
with patch('vllm_ascend.worker.model_runner_v1.get_ascend_soc_version',
return_value=unsupported_soc), \
patch('vllm_ascend.worker.model_runner_v1.is_global_first_rank',
return_value=True), \
pytest.raises(ValueError, match=f"Unsupported soc_version: {unsupported_soc}"):
NPUModelRunner._select_moe_comm_method(mock_runner, 100, False)

100
tests/ut/worker/test_worker_v1.py
View File

@@ -258,7 +258,7 @@ class TestNPUWorker(TestBase):
# Create worker mock
with patch.object(NPUWorker, "__init__", lambda x, **kwargs: None):
worker = NPUWorker()
worker._sleep_saved_buffers = {}
# Test wake_up method
worker.wake_up(tags=["test_tag"])
@@ -355,6 +355,28 @@ class TestNPUWorker(TestBase):
self.assertIn("Profiler is not enabled", str(cm.exception))
@patch("vllm_ascend.worker.worker_v1.envs_vllm")
@patch("vllm_ascend.worker.worker_v1.envs_ascend")
def test_profile_and_msmonitor_both_enabled_raises_error(
self, mock_envs_vllm, mock_envs_ascend):
"""Test profile method raises exception when both profiler and msmonitor are enabled"""
from vllm_ascend.worker.worker_v1 import NPUWorker
mock_envs_vllm.VLLM_TORCH_PROFILER_DIR = "/path/to/traces"
mock_envs_ascend.MSMONITOR_USE_DAEMON = 1
# Create worker mock
with patch.object(NPUWorker, "__init__", lambda x, **kwargs: None):
worker = NPUWorker()
# Test should raise exception
with self.assertRaises(RuntimeError) as cm:
_ = worker._init_profiler()
self.assertIn(
"MSMONITOR_USE_DAEMON and VLLM_TORCH_PROFILER_DIR cannot be both set at the same time.",
str(cm.exception))
def test_lora_methods(self):
"""Test LoRA related methods"""
from vllm_ascend.worker.worker_v1 import NPUWorker
@@ -828,6 +850,7 @@ class TestNPUWorker(TestBase):
# Mock scheduler_output and return result
mock_scheduler_output = MagicMock()
mock_scheduler_output.total_num_scheduled_tokens = 1
# Create a real ModelRunnerOutput instance or mock
mock_model_output = MagicMock(spec=ModelRunnerOutput)
worker.model_runner.execute_model.return_value = mock_model_output
@@ -842,9 +865,8 @@ class TestNPUWorker(TestBase):
@patch("vllm_ascend.worker.worker_v1.get_pp_group")
@patch("vllm_ascend.worker.worker_v1.get_tp_group")
@patch("vllm_ascend.worker.worker_v1.has_kv_transfer_group")
def test_execute_model_middle_rank(self, mock_has_kv_transfer_group,
mock_get_tp_group, mock_get_pp_group):
def test_execute_model_middle_rank(self, mock_get_tp_group,
mock_get_pp_group):
"""Test execute_model method - middle rank case"""
from vllm.sequence import IntermediateTensors
@@ -875,10 +897,8 @@ class TestNPUWorker(TestBase):
)
worker.model_runner.execute_model.return_value = mock_intermediate_output
# Set has_kv_transfer_group returns False
mock_has_kv_transfer_group.return_value = False
mock_scheduler_output = MagicMock()
mock_scheduler_output.total_num_scheduled_tokens = 1
# Test execute_model
result = worker.execute_model(mock_scheduler_output)
@@ -926,6 +946,7 @@ class TestNPUWorker(TestBase):
# Mock return result
mock_scheduler_output = MagicMock()
mock_scheduler_output.total_num_scheduled_tokens = 1
mock_model_output = MagicMock(spec=ModelRunnerOutput)
worker.model_runner.execute_model.return_value = mock_model_output
@@ -1009,7 +1030,9 @@ class TestNPUWorker(TestBase):
@patch("vllm_ascend.worker.worker_v1.NPUPlatform.seed_everything")
@patch("vllm_ascend.worker.worker_v1.logger")
def test_compile_or_warm_up_model_with_eager_mode(self, mock_logger,
@patch("vllm_ascend.worker.worker_v1.NPUWorker._warm_up_atb")
def test_compile_or_warm_up_model_with_eager_mode(self, mock_warm_up_atb,
mock_logger,
mock_seed_everything):
"""Test compile_or_warm_up_model method - eager mode"""
from vllm_ascend.worker.worker_v1 import NPUWorker
@@ -1051,10 +1074,14 @@ class TestNPUWorker(TestBase):
# Verify seed setting
mock_seed_everything.assert_called_once_with(12345)
# Verify atb warm up
mock_warm_up_atb.assert_called_once()
@patch("vllm_ascend.worker.worker_v1.NPUPlatform.seed_everything")
@patch("vllm_ascend.worker.worker_v1.logger")
@patch("vllm_ascend.worker.worker_v1.NPUWorker._warm_up_atb")
def test_compile_or_warm_up_model_with_graph_capture(
self, mock_logger, mock_seed_everything):
self, mock_warm_up_atb, mock_logger, mock_seed_everything):
"""Test compile_or_warm_up_model method - with graph capture enabled"""
from vllm_ascend.worker.worker_v1 import NPUWorker
@@ -1087,6 +1114,9 @@ class TestNPUWorker(TestBase):
# Verify seed setting
mock_seed_everything.assert_called_once_with(67890)
# Verify atb warm up
mock_warm_up_atb.assert_called_once()
@patch("vllm_ascend.worker.worker_v1.CaMemAllocator")
def test_initialize_from_config_with_sleep_mode(self,
mock_allocator_class):
@@ -1141,3 +1171,55 @@ class TestNPUWorker(TestBase):
# Verify calls
worker.model_runner.initialize_kv_cache.assert_called_once_with(
mock_kv_cache_config)
@patch("vllm_ascend.worker.worker_v1.get_pp_group")
@patch("vllm_ascend.worker.worker_v1.get_tp_group")
@patch("vllm_ascend.worker.worker_v1.EMPTY_MODEL_RUNNER_OUTPUT")
def test_execute_model_kv_connector_not_finished(self, mock_empty_output,
mock_get_tp_group,
mock_get_pp_group):
"""Test execute_model method - kv_connector_output not finished sending/recving case"""
from vllm.sequence import IntermediateTensors
from vllm_ascend.worker.worker_v1 import NPUWorker
# Create worker mock
with patch.object(NPUWorker, "__init__", lambda x, **kwargs: None):
worker = NPUWorker()
worker.model_runner = MagicMock()
worker.vllm_config = MagicMock()
worker.vllm_config.parallel_config = MagicMock()
worker.vllm_config.parallel_config.distributed_executor_backend = "ray"
# Set as middle rank (not first, not last)
mock_pp_group = MagicMock()
mock_pp_group.is_first_rank = False
mock_pp_group.is_last_rank = False
mock_get_pp_group.return_value = mock_pp_group
# Setup tensor reception data
mock_pp_group.recv_tensor_dict.return_value = {"tensor": "data"}
# Create mock kv_connector_output - both finished_sending and finished_recving are False
mock_kv_connector_output = MagicMock()
mock_kv_connector_output.finished_sending = False
mock_kv_connector_output.finished_recving = False
# Mock return IntermediateTensors with kv_connector_output
mock_intermediate_output = MagicMock(spec=IntermediateTensors)
mock_intermediate_output.tensors = {"output_tensor": "data"}
mock_intermediate_output.kv_connector_output = mock_kv_connector_output
worker.model_runner.execute_model.return_value = mock_intermediate_output
mock_scheduler_output = MagicMock()
mock_scheduler_output.total_num_scheduled_tokens = 1
# Test execute_model
result = worker.execute_model(mock_scheduler_output)
# Verify tensor reception and sending
mock_pp_group.recv_tensor_dict.assert_called_once()
mock_pp_group.send_tensor_dict.assert_called_once()
# When both finished_sending and finished_recving are False, should return EMPTY_MODEL_RUNNER_OUTPUT directly
self.assertEqual(result, mock_empty_output)