Skip to content

Fix some ci issue and refactor modelrunner #2445

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
wangxiyuan merged 6 commits into from
Aug 20, 2025
Merged

Fix some ci issue and refactor modelrunner #2445

Show file tree
Hide file tree
Changes from 5 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
2fcd79a
support logitsprocessor
Potabk Aug 18, 2025
1ed0790
Some fixes to adapt latest vllm
MengqingCao Aug 18, 2025
3fcf35a
refact attention metadata build
weiguihua2 Aug 19, 2025
128f00a
refact attention metadata build
weiguihua2 Aug 19, 2025
54582c7
revert ci change
MengqingCao Aug 19, 2025
b0de043
fix ci
MengqingCao Aug 19, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 11 additions & 11 deletions .github/workflows/vllm_ascend_test.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,16 +40,16 @@ concurrency:
cancel-in-progress: true

jobs:
lint:
uses: ./.github/workflows/pre-commit.yml
# lint:
# uses: ./.github/workflows/pre-commit.yml

changes:
runs-on: ubuntu-latest
outputs:
e2e_tracker: ${{ steps.filter.outputs.e2e_tracker }}
ut_tracker: ${{ steps.filter.outputs.ut_tracker }}
steps:
- uses: actions/checkout@v5
- uses: actions/checkout@v4
- uses: dorny/paths-filter@v3
id: filter
with:
Expand Down Expand Up @@ -130,9 +130,9 @@ jobs:
verbose: true

e2e:
needs: [lint, changes]
needs: [changes]
# only trigger e2e test after lint passed and the change is e2e related with pull request.
if: ${{ github.event_name == 'pull_request' && needs.lint.result == 'success' && needs.changes.outputs.e2e_tracker == 'true' }}
if: ${{ github.event_name == 'pull_request' && needs.changes.outputs.e2e_tracker == 'true' }}
strategy:
max-parallel: 2
matrix:
Expand Down Expand Up @@ -160,15 +160,15 @@ jobs:
apt install git -y

- name: Checkout vllm-project/vllm-ascend repo
uses: actions/checkout@v5
uses: actions/checkout@v4

- name: Install system dependencies
run: |
apt-get -y install `cat packages.txt`
apt-get -y install gcc g++ cmake libnuma-dev

- name: Checkout vllm-project/vllm repo
uses: actions/checkout@v5
uses: actions/checkout@v4
with:
repository: vllm-project/vllm
ref: ${{ matrix.vllm_version }}
Expand All @@ -192,7 +192,7 @@ jobs:
VLLM_USE_MODELSCOPE: True
run: |
pytest -sv tests/e2e/singlecard/test_offline_inference.py
pytest -sv tests/e2e/singlecard/test_ilama_lora.py
# pytest -sv tests/e2e/singlecard/test_ilama_lora.py
pytest -sv tests/e2e/singlecard/test_guided_decoding.py
pytest -sv tests/e2e/singlecard/test_camem.py
pytest -sv tests/e2e/singlecard/test_embedding.py
Expand Down Expand Up @@ -242,15 +242,15 @@ jobs:
apt install git -y

- name: Checkout vllm-project/vllm-ascend repo
uses: actions/checkout@v5
uses: actions/checkout@v4

- name: Install system dependencies
run: |
apt-get -y install `cat packages.txt`
apt-get -y install gcc g++ cmake libnuma-dev

- name: Checkout vllm-project/vllm repo
uses: actions/checkout@v5
uses: actions/checkout@v4
with:
repository: vllm-project/vllm
ref: ${{ matrix.vllm_version }}
Expand All @@ -273,7 +273,7 @@ jobs:
VLLM_WORKER_MULTIPROC_METHOD: spawn
VLLM_USE_MODELSCOPE: True
run: |
pytest -sv tests/e2e/multicard/test_ilama_lora_tp2.py
# pytest -sv tests/e2e/multicard/test_ilama_lora_tp2.py
# Fixme: run VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/multicard/test_offline_inference_distributed.py will raise error.
# To avoid oom, we need to run the test in a single process.
pytest -sv tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_DeepSeek_multistream_moe
Expand Down
2 changes: 1 addition & 1 deletion examples/offline_inference_audio_language.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@

from vllm.assets.audio import AudioAsset
try:
import librosa
import librosa # type: ignore
except ImportError:
raise Exception("Can't import librosa, please ensure it's installed")

Expand Down
114 changes: 59 additions & 55 deletions tests/ut/attention/test_attention_v1.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,7 @@
AscendAttentionState,
AscendMetadata,
CommonAttentionState)
from vllm_ascend.attention.utils import AscendCommonAttentionMetadata


class TestAscendAttentionBackend(TestBase):
Expand Down Expand Up @@ -67,8 +68,12 @@ def test_copy_blocks(self):
class TestAscendAttentionMetadataBuilder(TestBase):

def setUp(self):
self.mock_runner = MagicMock()
self.builder = AscendAttentionMetadataBuilder(self.mock_runner)
self.mock_vllm_config = MagicMock()
self.mock_vllm_config.model_config.max_model_len = 640
self.mock_vllm_config.cache_config.block_size = 64
self.mock_device = 'cpu:0'
self.builder = AscendAttentionMetadataBuilder(self.mock_vllm_config,
self.mock_device)

def test_reorder_batch(self):
mock_input_batch = MagicMock()
Expand All @@ -86,31 +91,28 @@ def test_reorder_batch(self):
def test_build_prefill_no_cache(self, mock_is_310p, mock_nd_to_nz_2d,
mock_npu_format_cast,
mock_ascend_metadata):
num_reqs = 2
num_actual_tokens = 10
max_query_len = 5

self.mock_runner.input_batch.block_table = [MagicMock()]
self.mock_runner.input_batch.block_table[
0].get_device_tensor.return_value = torch.zeros((10, 10))
self.mock_runner.max_num_blocks_per_req = 10
self.mock_runner.query_lens = torch.tensor([3, 4])
self.mock_runner.seq_lens_cpu = torch.tensor([5, 6])
self.mock_runner.slot_mapping_cpu = torch.tensor(range(20))
self.mock_runner.device = 'cpu:0'
self.mock_runner.attn_mask = torch.ones((10, 10))
self.mock_runner.attn_state = AscendAttentionState.PrefillNoCache
self.mock_runner.query_start_loc_cpu = torch.tensor([0, 3, 7])
common_attn_metadata = AscendCommonAttentionMetadata(
query_start_loc=torch.tensor([0, 3, 7]),
query_start_loc_cpu=torch.tensor([0, 3, 7]),
seq_lens_cpu=torch.tensor([5, 6]),
num_reqs=2,
num_actual_tokens=10,
max_query_len=5,
decode_token_per_req=torch.tensor([1, 1]),
block_table_tensor=torch.zeros((10, 10)),
slot_mapping_cpu=torch.tensor(range(20)),
actual_seq_lengths_q=torch.tensor([0, 1]),
positions=torch.tensor([10, 10]),
attn_mask=torch.ones((10, 10)),
spec_attn_mask=None,
attn_state=AscendAttentionState.PrefillNoCache)

mock_nz_tensor = MagicMock()
mock_model = MagicMock()
mock_nd_to_nz_2d.return_value = mock_nz_tensor
mock_npu_format_cast.return_value = mock_nz_tensor

self.builder.build(
num_reqs,
num_actual_tokens,
max_query_len,
)
self.builder.build(common_attn_metadata, mock_model)

@patch('vllm_ascend.attention.attention_v1.AscendMetadata')
@patch('torch_npu.npu_format_cast')
Expand All @@ -120,51 +122,53 @@ def test_build_prefill_no_cache(self, mock_is_310p, mock_nd_to_nz_2d,
def test_build_chunked_prefill(self, mock_ascend_attention_state,
mock_is_310p, mock_nd_to_nz_spec,
mock_npu_format_cast, mock_ascend_metadata):
num_reqs = 3
num_actual_tokens = 15
max_query_len = 6

self.mock_runner.input_batch.block_table = [MagicMock()]
self.mock_runner.input_batch.block_table[
0].get_device_tensor.return_value = torch.zeros((10, 10))
self.mock_runner.max_num_blocks_per_req = 10
self.mock_runner.query_lens = torch.tensor([2, 3, 4])
self.mock_runner.seq_lens_cpu = torch.tensor([4, 5, 6])
self.mock_runner.slot_mapping_cpu = torch.tensor(range(20))
self.mock_runner.device = 'cpu:0'
self.mock_runner.attn_mask = torch.ones((15, 15))
self.mock_runner.attn_state = AscendAttentionState.ChunkedPrefill
self.mock_runner.query_start_loc_cpu = torch.tensor([0, 2, 5, 9])
common_attn_metadata = AscendCommonAttentionMetadata(
query_start_loc=torch.tensor([0, 2, 5, 9]),
query_start_loc_cpu=torch.tensor([0, 2, 5, 9]),
seq_lens_cpu=torch.tensor([4, 5, 6]),
num_reqs=3,
num_actual_tokens=15,
max_query_len=6,
decode_token_per_req=torch.tensor([1, 1, 1]),
block_table_tensor=torch.zeros((10, 10)),
slot_mapping_cpu=torch.tensor(range(20)),
actual_seq_lengths_q=torch.tensor([0, 1, 2]),
positions=torch.tensor([10, 10]),
attn_mask=torch.ones((15, 15)),
spec_attn_mask=None,
attn_state=AscendAttentionState.ChunkedPrefill)

mock_ascend_attention_state = MagicMock()
mock_ascend_attention_state.PrefillNoCache = 0

mock_nz_tensor = MagicMock()
mock_model = MagicMock()
mock_nd_to_nz_spec.return_value = mock_nz_tensor
mock_npu_format_cast.return_value = mock_nz_tensor

self.builder.build(num_reqs, num_actual_tokens, max_query_len)
self.builder.build(common_attn_metadata, mock_model)

@patch('vllm_ascend.attention.attention_v1.AscendMetadata')
@patch('vllm_ascend.attention.attention_v1.is_310p', return_value=False)
def test_build_non_310p(self, mock_is_310p, mock_ascend_metadata):
num_reqs = 3
num_actual_tokens = 15
max_query_len = 6

self.mock_runner.input_batch.block_table = [MagicMock()]
self.mock_runner.input_batch.block_table[
0].get_device_tensor.return_value = torch.zeros((10, 10))
self.mock_runner.max_num_blocks_per_req = 10
self.mock_runner.query_lens = torch.tensor([2, 3, 4])
self.mock_runner.seq_lens_cpu = torch.tensor([4, 5, 6])
self.mock_runner.slot_mapping_cpu = torch.tensor(range(20))
self.mock_runner.device = 'cpu:0'
self.mock_runner.attn_mask = torch.ones((15, 15))
self.mock_runner.attn_state = AscendAttentionState.ChunkedPrefill
self.mock_runner.query_start_loc_cpu = torch.tensor([0, 2, 5, 9])

self.builder.build(num_reqs, num_actual_tokens, max_query_len)
common_attn_metadata = AscendCommonAttentionMetadata(
query_start_loc=torch.tensor([0, 2, 5, 9]),
query_start_loc_cpu=torch.tensor([0, 2, 5, 9]),
seq_lens_cpu=torch.tensor([4, 5, 6]),
num_reqs=3,
num_actual_tokens=15,
max_query_len=6,
decode_token_per_req=torch.tensor([1, 1, 1]),
block_table_tensor=torch.zeros((10, 10)),
slot_mapping_cpu=torch.tensor(range(20)),
actual_seq_lengths_q=torch.tensor([0, 1, 2]),
positions=torch.tensor([10, 10]),
attn_mask=torch.ones((15, 15)),
spec_attn_mask=None,
attn_state=AscendAttentionState.ChunkedPrefill)
mock_model = MagicMock()

self.builder.build(common_attn_metadata, mock_model)


class TestAscendAttentionBackendImpl(TestBase):
Expand Down
Loading
Loading