Skip to content

Commit e9359bd

Browse files
wangxiyuanwangxiyuan
authored
[CI] Pin vLLM to releases/v0.11.0 (#3211)
### What this PR does / why we need it? - Pin vLLM commit to releases/v0.11.0 branch. - Fix the break change by vLLM commit vllm-project/vllm@d4d9899 ### Does this PR introduce _any_ user-facing change? no ### How was this patch tested? - vLLM version: v0.10.2 - vLLM main: vllm-project/vllm@17b4c66 Signed-off-by: wangxiyuan <[email protected]>
1 parent 9caf6fb commit e9359bd

File tree

6 files changed

+83
-38
lines changed

6 files changed

+83
-38
lines changed

‎.github/workflows/format_pr_body.yaml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -36,7 +36,7 @@ jobs:
3636

3737
- name: Get vLLM version
3838
run: |
39-
VLLM_COMMIT=17b4c6685ce62d5652654784d6771a3d38e4273e
39+
VLLM_COMMIT=releases/v0.11.0
4040
echo "VLLM_COMMIT=https://github.com/vllm-project/vllm/commit/$VLLM_COMMIT" >> $GITHUB_ENV
4141
4242
- name: Checkout repository

‎.github/workflows/vllm_ascend_test.yaml‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -42,7 +42,7 @@ jobs:
4242
lint:
4343
uses: ./.github/workflows/pre-commit.yml
4444
with:
45-
vllm: 17b4c6685ce62d5652654784d6771a3d38e4273e
45+
vllm: releases/v0.11.0
4646

4747
changes:
4848
runs-on: ubuntu-latest
@@ -83,7 +83,7 @@ jobs:
8383
VLLM_USE_MODELSCOPE: True
8484
strategy:
8585
matrix:
86-
vllm_version: [17b4c6685ce62d5652654784d6771a3d38e4273e, v0.10.2]
86+
vllm_version: [releases/v0.11.0, v0.10.2]
8787
steps:
8888
- name: Install packages
8989
run: |
@@ -138,7 +138,7 @@ jobs:
138138
name: e2e-light
139139
strategy:
140140
matrix:
141-
vllm_version: [17b4c6685ce62d5652654784d6771a3d38e4273e, v0.10.2]
141+
vllm_version: [releases/v0.11.0, v0.10.2]
142142
# Note (yikun): If CI resource are limited we can split job into two chain jobs
143143
needs: [lint, changes]
144144
# only trigger e2e test after lint passed and the change is e2e related with pull request.

‎.github/workflows/vllm_ascend_test_full.yaml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -68,7 +68,7 @@ jobs:
6868
name: e2e-full
6969
strategy:
7070
matrix:
71-
vllm_version: [17b4c6685ce62d5652654784d6771a3d38e4273e, v0.10.2]
71+
vllm_version: [releases/v0.11.0, v0.10.2]
7272
needs: [changes]
7373
if: ${{ needs.changes.outputs.e2e_tracker == 'true' }}
7474
uses: ./.github/workflows/_e2e_test.yaml

‎vllm_ascend/models/qwen2_5_vl.py‎

Lines changed: 16 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -42,6 +42,8 @@
4242
from vllm.model_executor.models.utils import maybe_prefix
4343
from vllm.multimodal import MULTIMODAL_REGISTRY
4444

45+
from vllm_ascend.utils import vllm_version_is
46+
4547
MIN_PAD_SIZE = 64 # min_size to pad weight
4648
MAX_PAD_SIZE = 128 # max_size to pad weight
4749

@@ -496,12 +498,20 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
496498
super().__init__(vllm_config=vllm_config, prefix=prefix)
497499
config: Qwen2_5_VLConfig = vllm_config.model_config.hf_config
498500
quant_config = vllm_config.quant_config
499-
self.visual = AscendQwen2_5_VisionTransformer(
500-
vision_config=config.vision_config,
501-
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
502-
quant_config=self._maybe_ignore_quant_config(quant_config),
503-
prefix=maybe_prefix(prefix, "visual"),
504-
)
501+
if vllm_version_is("0.10.2"):
502+
self.visual = AscendQwen2_5_VisionTransformer(
503+
vision_config=config.vision_config,
504+
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
505+
quant_config=self._maybe_ignore_quant_config(quant_config),
506+
prefix=maybe_prefix(prefix, "visual"),
507+
)
508+
else:
509+
self.visual = AscendQwen2_5_VisionTransformer(
510+
vision_config=config.vision_config,
511+
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
512+
quant_config=self.quant_config,
513+
prefix=maybe_prefix(prefix, "visual"),
514+
)
505515

506516
def _process_image_input(self, image_input) -> tuple[torch.Tensor, ...]:
507517

‎vllm_ascend/models/qwen2_5_vl_without_padding.py‎

Lines changed: 45 additions & 20 deletions
Original file line numberDiff line numberDiff line change
@@ -68,6 +68,7 @@
6868
from vllm.multimodal import MULTIMODAL_REGISTRY
6969

7070
from vllm_ascend.models.qwen2_5_vl import AscendQwen2_5_VisionRotaryEmbedding
71+
from vllm_ascend.utils import vllm_version_is
7172

7273

7374
class AscendQwen2_5_VisionAttention_Without_Padding(Qwen2_5_VisionAttention):
@@ -483,12 +484,20 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
483484
super().__init__(vllm_config=vllm_config, prefix=prefix)
484485
config: Qwen2_5_VLConfig = vllm_config.model_config.hf_config
485486
quant_config = vllm_config.quant_config
486-
self.visual = AscendQwen2_5_VisionTransformer_Without_Padding(
487-
vision_config=config.vision_config,
488-
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
489-
quant_config=self._maybe_ignore_quant_config(quant_config),
490-
prefix=maybe_prefix(prefix, "visual"),
491-
)
487+
if vllm_version_is("0.10.2"):
488+
self.visual = AscendQwen2_5_VisionTransformer_Without_Padding(
489+
vision_config=config.vision_config,
490+
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
491+
quant_config=self._maybe_ignore_quant_config(quant_config),
492+
prefix=maybe_prefix(prefix, "visual"),
493+
)
494+
else:
495+
self.visual = AscendQwen2_5_VisionTransformer_Without_Padding(
496+
vision_config=config.vision_config,
497+
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
498+
quant_config=self.quant_config,
499+
prefix=maybe_prefix(prefix, "visual"),
500+
)
492501

493502
def _process_image_input(self, image_input) -> tuple[torch.Tensor, ...]:
494503

@@ -554,12 +563,20 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
554563
super().__init__(vllm_config=vllm_config, prefix=prefix)
555564
config: Qwen3VLConfig = vllm_config.model_config.hf_config
556565
quant_config = vllm_config.quant_config
557-
self.visual = AscendQwen3_VisionTransformer(
558-
config.vision_config,
559-
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
560-
quant_config=self._maybe_ignore_quant_config(quant_config),
561-
prefix=maybe_prefix(prefix, "visual"),
562-
use_data_parallel=self.use_data_parallel)
566+
if vllm_version_is("0.10.2"):
567+
self.visual = AscendQwen3_VisionTransformer(
568+
config.vision_config,
569+
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
570+
quant_config=self._maybe_ignore_quant_config(quant_config),
571+
prefix=maybe_prefix(prefix, "visual"),
572+
use_data_parallel=self.use_data_parallel)
573+
else:
574+
self.visual = AscendQwen3_VisionTransformer(
575+
config.vision_config,
576+
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
577+
quant_config=self.quant_config,
578+
prefix=maybe_prefix(prefix, "visual"),
579+
use_data_parallel=self.use_data_parallel)
563580

564581

565582
@MULTIMODAL_REGISTRY.register_processor(Qwen3VLMultiModalProcessor,
@@ -596,11 +613,19 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
596613
multimodal_config = vllm_config.model_config.multimodal_config
597614
self.multimodal_config = multimodal_config
598615
self.use_data_parallel = multimodal_config.mm_encoder_tp_mode == "data"
599-
600-
self.visual = AscendQwen3_VisionTransformer(
601-
config.vision_config,
602-
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
603-
quant_config=self._maybe_ignore_quant_config(quant_config),
604-
prefix=maybe_prefix(prefix, "visual"),
605-
use_data_parallel=self.use_data_parallel,
606-
)
616+
if vllm_version_is("0.10.2"):
617+
self.visual = AscendQwen3_VisionTransformer(
618+
config.vision_config,
619+
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
620+
quant_config=self._maybe_ignore_quant_config(quant_config),
621+
prefix=maybe_prefix(prefix, "visual"),
622+
use_data_parallel=self.use_data_parallel,
623+
)
624+
else:
625+
self.visual = AscendQwen3_VisionTransformer(
626+
config.vision_config,
627+
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
628+
quant_config=self.quant_config,
629+
prefix=maybe_prefix(prefix, "visual"),
630+
use_data_parallel=self.use_data_parallel,
631+
)

‎vllm_ascend/models/qwen2_vl.py‎

Lines changed: 17 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -40,6 +40,8 @@
4040
from vllm.model_executor.models.utils import maybe_prefix
4141
from vllm.multimodal import MULTIMODAL_REGISTRY
4242

43+
from vllm_ascend.utils import vllm_version_is
44+
4345
MIN_PAD_SIZE = 64 # min_size to pad weight
4446
MAX_PAD_SIZE = 128 # max_size to pad weight
4547

@@ -343,10 +345,18 @@ class AscendQwen2VLForConditionalGeneration(Qwen2VLForConditionalGeneration):
343345

344346
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
345347
super().__init__(vllm_config=vllm_config, prefix=prefix)
346-
self.visual = AscendQwen2VisionTransformer(
347-
self.config.vision_config,
348-
norm_eps=getattr(self.config, "rms_norm_eps", 1e-6),
349-
quant_config=self._maybe_ignore_quant_config(
350-
vllm_config.quant_config),
351-
prefix=maybe_prefix(prefix, "visual"),
352-
)
348+
if vllm_version_is("0.10.2"):
349+
self.visual = AscendQwen2VisionTransformer(
350+
self.config.vision_config,
351+
norm_eps=getattr(self.config, "rms_norm_eps", 1e-6),
352+
quant_config=self._maybe_ignore_quant_config(
353+
vllm_config.quant_config),
354+
prefix=maybe_prefix(prefix, "visual"),
355+
)
356+
else:
357+
self.visual = AscendQwen2VisionTransformer(
358+
self.config.vision_config,
359+
norm_eps=getattr(self.config, "rms_norm_eps", 1e-6),
360+
quant_config=self.vllm_config.quant_config,
361+
prefix=maybe_prefix(prefix, "visual"),
362+
)

0 commit comments

Comments
 (0)