frozenleaves/vllm-ascend
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm-ascend.git synced 2025-10-20 21:53:54 +08:00
Code Issues Packages Projects Releases Wiki Activity

[main][bugfix] bugfix for minicpm models (#3527)

Browse Source 
### What this PR does / why we need it?
bugfix for minicpm-2b and minicpm3-4b

- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

Signed-off-by: Wang Kunpeng <1289706727@qq.com>
This commit is contained in:
Wang Kunpeng
2025-10-19 11:00:55 +08:00
committed by GitHub
parent 6c9909c861
commit 4b3bd4f397
3 changed files with 3 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
.github/workflows/vllm_ascend_test.yaml vendored
View File

@ -121,7 +121,6 @@ jobs:
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/x86_64-linux/devlib
pytest -sv --cov --cov-report=xml:unittests-coverage.xml tests/ut \
--ignore=tests/ut/test_platform.py \
--ignore=tests/ut/patch/worker/patch_common/test_patch_minicpm.py \
--ignore=tests/ut/core/test_scheduler.py \
--ignore=tests/ut/kv_connector/test_llmdatadist_connector.py \
--ignore=tests/ut/kv_connector/test_mooncake_connector.py \

4
vllm_ascend/patch/worker/patch_common/__init__.py
View File

@ -26,6 +26,4 @@ import vllm_ascend.patch.worker.patch_common.patch_logits # noqa
import vllm_ascend.patch.worker.patch_common.patch_roberta # noqa
import vllm_ascend.patch.worker.patch_common.patch_weight_loader # noqa
import vllm_ascend.patch.worker.patch_common.patch_multimodal_merge # noqa
# TODO: revert me when triton import is fixed
# import vllm_ascend.patch.worker.patch_common.patch_minicpm # noqa
import vllm_ascend.patch.worker.patch_common.patch_minicpm # noqa

2
vllm_ascend/worker/model_runner_v1.py
View File

@ -1346,6 +1346,8 @@ class NPUModelRunner(LoRAModelRunnerMixin):
positions_cpu = self.positions_cpu[:num_input_tokens]
positions = self.positions[:num_input_tokens]
seq_lens_cpu = self.seq_lens_cpu[:num_reqs]
attn_state = self._build_attn_state(num_reqs, num_scheduled_tokens,
num_valid_tokens)
self.attn_mask = self._make_attention_mask(seq_lens=seq_lens_cpu,
position=positions_cpu,
attn_state=attn_state)