EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[CI] Upgrade vLLM to 20250919 (6d8246aa) and fix some broken issue (#2907)

Browse Source 
### What this PR does / why we need it?
1. This pr bump vllm commit to
6d8246aaff
2. fix upstream changes https://github.com/vllm-project/vllm/pull/24548
abort multi-modal kwargs, make vllm main and `v0.10.2` both adaptable
3. fix metadata_builder changes introduced by
https://github.com/vllm-project/vllm/pull/23693
4. fix `structured_outputs_config` changes introduced by
https://github.com/vllm-project/vllm/pull/22772
5. fix `moe_config` changes introduced by
https://github.com/vllm-project/vllm/pull/22537

Co-authored-by:  MengqingCao <cmq0113@163.com>
Co-authored-by:  Yikun Jiang <yikunkero@gmail.com>


- vLLM version: v0.10.2
- vLLM main:
c60e6137f0

---------

Signed-off-by: wangli <wangli858794774@gmail.com>
Signed-off-by: MengqingCao <cmq0113@163.com>
Co-authored-by: MengqingCao <cmq0113@163.com>
This commit is contained in:
Li Wang
2025-09-20 17:37:57 +08:00
committed by GitHub
parent 53ecd89e8f
commit 12bcbd02bb
14 changed files with 359 additions and 143 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
vllm_ascend/ops/fused_moe.py
View File

@@ -47,7 +47,8 @@ from vllm_ascend.ops.moe.moe_comm_method import (AllGatherCommImpl,
from vllm_ascend.ops.sequence_parallel import MetadataForPadding
from vllm_ascend.utils import (ACL_FORMAT_FRACTAL_NZ,
get_all_reduce_merge_state,
get_rm_router_logits_state, is_310p)
get_rm_router_logits_state, is_310p,
vllm_version_is)
class AscendUnquantizedFusedMoEMethod(UnquantizedFusedMoEMethod):
@@ -278,16 +279,25 @@ class AscendFusedMoE(FusedMoE):
if self.scoring_func != "softmax" and not self.use_grouped_topk:
raise ValueError("Only softmax scoring function is supported for "
"non-grouped topk.")
moe = FusedMoEConfig.make(
num_experts=self.global_num_experts,
experts_per_token=top_k,
hidden_dim=hidden_size,
num_local_experts=self.local_num_experts,
moe_parallel_config=self.moe_parallel_config,
# TODO (bnell): this needs to be fixed for quantized types.
in_dtype=params_dtype,
quant_config=quant_config)
if vllm_version_is("0.10.2"):
moe = FusedMoEConfig.make(
num_experts=self.global_num_experts,
experts_per_token=top_k,
hidden_dim=hidden_size,
num_local_experts=self.local_num_experts,
moe_parallel_config=self.moe_parallel_config,
# TODO (bnell): this needs to be fixed for quantized types.
in_dtype=params_dtype,
quant_config=quant_config)
else:
moe = FusedMoEConfig(
num_experts=self.global_num_experts,
experts_per_token=top_k,
hidden_dim=hidden_size,
num_local_experts=self.local_num_experts,
moe_parallel_config=self.moe_parallel_config,
in_dtype=params_dtype,
)
self.moe_config = moe
if quant_config is None: