[Bugfix] Follow vLLM Qwen-Moe/VL and KV Connector change to fix broken CI (#2181)

### What this PR does / why we need it?
This pr fix broken CI:
1. Fix the
ee2eb6ecd8
changes, in this commit, they fused the gate and up projections in the
vision MLP, This can improve performance by reducing one matrix
multiplication. so, this pr do the following things:
- Specify that the two linear layers are fused as `mlp.gate_up_proj`
when loading the weights.
    - Use a SiluAndMul activation function.
2. Fix
aefeea0fde,
Update ModelRunnerOutput parameters to adapt to its changes
3. Fix
[vllm-commit](https://github.com/vllm-project/vllm/pull/20815/files#diff-3ffb829a39ab2b3e4706aa28f5e476815f36c3a87b98d6a66514ebedc8f3ffb4R354-R356),
fix qwen moe
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?


- vLLM version: v0.10.0
- vLLM main:
fed5849d3f

---------

Signed-off-by: wangli <wangli858794774@gmail.com>
This commit is contained in:
Li Wang
2025-08-04 21:37:50 +08:00
committed by GitHub
parent e38fab011d
commit ad366bf908
8 changed files with 137 additions and 56 deletions

View File

@@ -186,6 +186,20 @@ def create_model_runner_output(
sampled_token_ids = [[sampled_token] for _ in req_ids]
# Make output data structure.
extra_args = {}
if not vllm_version_is("0.10.0"):
from vllm.v1.worker.kv_connector_model_runner_mixin import \
KVConnectorOutput # type: ignore # noqa
kv_connector_output = KVConnectorOutput(
finished_sending=finished_sending,
finished_recving=finished_recving)
extra_args = {"kv_connector_output": kv_connector_output}
else:
extra_args = {
"finished_sending": finished_sending,
"finished_recving": finished_recving,
}
return ModelRunnerOutput(
req_ids=req_ids,
req_id_to_index=req_id_to_index,
@@ -193,9 +207,6 @@ def create_model_runner_output(
spec_token_ids=None,
logprobs=None,
prompt_logprobs_dict={},
**({
"pooler_output": []
} if not vllm_version_is("0.9.1") else {}),
finished_sending=finished_sending,
finished_recving=finished_recving,
pooler_output=[],
**extra_args,
)