[Main2Main] Upgrade vllm commit to 0113 (#5839)

### What this PR does / why we need it?
Upgrade vllm commit to 0113 (11b6af5280d6d6dfb8953af16e67b25f819b3be9)

- Modify import paths due to the refactors
https://github.com/vllm-project/vllm/pull/31916
https://github.com/vllm-project/vllm/pull/32054

- Fix `TypeError: NPUOffloadingSpec.__init__() takes 2 positional
arguments but 3 were given` due to
https://github.com/vllm-project/vllm/pull/24498

- Skip the async-scheduling tests in
`tests/e2e/multicard/4-cards/long_sequence/test_mtp.py`, which are never
verified
https://github.com/vllm-project/vllm/pull/31998

- Skip some pooling tests, which are caused by
https://github.com/vllm-project/vllm/pull/32148
where vllm is also failed
https://buildkite.com/vllm/ci/builds/46705/steps/canvas?jid=019bb329-3834-4685-862b-1613b8e0f5d4

We will reopen those tests when main2main reachs
https://github.com/vllm-project/vllm/pull/32243

- Skip some cases in
`tests/e2e/multicard/4-cards/long_sequence/test_mtp.py`, which are
broken by
https://github.com/vllm-project/vllm/pull/32118

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef

Signed-off-by: wjunLu <wjunlu217@gmail.com>
Signed-off-by: hfadzxy <starmoon_zhang@163.com>
Co-authored-by: hfadzxy <starmoon_zhang@163.com>
This commit is contained in:
wjunLu
2026-01-15 09:48:53 +08:00
committed by GitHub
parent e67608041d
commit c11a05c4e1
29 changed files with 229 additions and 54 deletions

View File

@@ -18,8 +18,10 @@
#
import os
import pytest
from tests.e2e.conftest import VllmRunner
from vllm_ascend.utils import vllm_version_is
os.environ["HCCL_BUFFSIZE"] = "512"
@@ -44,10 +46,15 @@ def test_pcp_dcp_mtp1_eager():
"method": "deepseek_mtp",
},
enforce_eager=True,
async_scheduling=False,
) as runner:
runner.generate_greedy(prompts, 32)
@pytest.mark.skipif(
not vllm_version_is('0.13.0'),
reason="vLLM PR-32118 break this",
)
def test_pcp_dcp_mtp3_eager():
prompts = [
"The capital of France is", "Hello, my name is Tom, I am",
@@ -68,10 +75,15 @@ def test_pcp_dcp_mtp3_eager():
"method": "deepseek_mtp",
},
enforce_eager=True,
async_scheduling=False,
) as runner:
runner.generate_greedy(prompts, 32)
@pytest.mark.skipif(
not vllm_version_is('0.13.0'),
reason="vLLM PR-32118 break this",
)
def test_pcp_dcp_mtp3_piecewise_graph():
prompts = [
"The capital of France is", "Hello, my name is Tom, I am",
@@ -95,10 +107,15 @@ def test_pcp_dcp_mtp3_piecewise_graph():
"cudagraph_mode": "PIECEWISE",
"cudagraph_capture_sizes": [4, 8, 16],
},
async_scheduling=False,
) as runner:
runner.generate_greedy(prompts, 32)
@pytest.mark.skipif(
not vllm_version_is('0.13.0'),
reason="vLLM PR-32118 break this",
)
def test_pcp_dcp_mtp3_full_graph():
prompts = [
"The capital of France is", "Hello, my name is Tom, I am",
@@ -122,6 +139,7 @@ def test_pcp_dcp_mtp3_full_graph():
"cudagraph_mode": "FULL_DECODE_ONLY",
"cudagraph_capture_sizes": [4, 8, 16],
},
async_scheduling=False,
) as runner:
runner.generate_greedy(prompts, 32)
@@ -148,5 +166,6 @@ def test_dcp_mtp3_full_graph():
"cudagraph_mode": "FULL_DECODE_ONLY",
"cudagraph_capture_sizes": [4, 8, 16],
},
async_scheduling=False,
) as runner:
runner.generate_greedy(prompts, 32)

View File

@@ -79,7 +79,7 @@ def test_qwen3_next_mtp_acceptance_tp4(model_name):
for num_accepted_tokens in num_accepted_tokens_per_pos
]
match = all(abs(a - b) < 0.06 for a, b in zip(acceptance_per_pos, golden))
match = all((a >= b) or (b - a < 0.06) for a, b in zip(acceptance_per_pos, golden))
if not match:
print(f"acceptance_per_pos: {acceptance_per_pos}")
print(f"golden: {golden}")