Files
xc-llm-ascend/tests/ut/worker/test_model_runner_v1.py
realliujiaxu 29bd9235ed [v0.11.0][Perf] Delete redundant operations in model_runner and forward_context (#3775)
<!--  Thanks for sending a pull request!

BEFORE SUBMITTING, PLEASE READ
https://docs.vllm.ai/en/latest/contributing/overview.html

-->

cherry pick https://github.com/vllm-project/vllm-ascend/pull/3677

Remove redundant operations from `model_runner` and `forward_context`.
This optimization can significantly reduce the idle time (bubble) before
decoding when running models with small parameter counts (e.g.,
Qwen/Qwen2.5-0.5B).

Testing on 800I A2, bubble is reduced from 3.8ms to 2.8ms :
Before
<img width="1655" height="696" alt="image"
src="https://github.com/user-attachments/assets/d7608e52-2438-46dd-8fc9-391fd6274495"
/>

After
<img width="1607" height="774" alt="image"
src="https://github.com/user-attachments/assets/56daf081-2dba-4d2e-99d4-e055187d9806"
/>
### What this PR does / why we need it?
<!--
- Please clarify what changes you are proposing. The purpose of this
section is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster
reviews in your PR.

- Please clarify why the changes are needed. For instance, the use case
and bug description.

- Fixes #
-->

### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such
as API, interface or other behavior changes.
Documentation-only updates are not considered user-facing changes.
-->
No
### How was this patch tested?
<!--
CI passed with new added/existing test.
If it was tested in a way different from regular unit tests, please
clarify how you tested step by step, ideally copy and paste-able, so
that other reviewers can test and check, and descendants can verify in
the future.
If tests were not added, please describe why they were not added and/or
why it was difficult to add.
-->

---------

Signed-off-by: realliujiaxu <realliujiaxu@163.com>
2025-10-29 15:58:53 +08:00

112 lines
4.8 KiB
Python

# 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), \
patch('vllm_ascend.worker.model_runner_v1.is_moe_model',
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), \
patch('vllm_ascend.worker.model_runner_v1.is_moe_model',
return_value=True), \
pytest.raises(ValueError, match=f"Unsupported soc_version: {unsupported_soc}"):
NPUModelRunner._select_moe_comm_method(mock_runner, 100, False)