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

[BugFix]Disable dispatch_gmm_combine_decode operator when mtp drafter model uses non-w8a8 while main model uses w8a8, or drafter model is eagle series (#5293)

Browse Source 
…w8a8 while main model uses w8a8

### What this PR does / why we need it?

Disable dispatch_gmm_combine_decode operator when mtp drafter model uses
non-w8a8 while main model uses w8a8, or drafter model is eagle series.

More info about this operator, please refer to RFC: issue
https://github.com/vllm-project/vllm-ascend/issues/5476


- vLLM version: release/v0.13.0
- vLLM main:
ad32e3e19c

Signed-off-by: wangqiankun <wangqiankun13@huawei.com>
This commit is contained in:
wangqiankun13
2026-01-04 17:51:28 +08:00
committed by GitHub
parent f15dc3fa02
commit 350b95efcf
2 changed files with 23 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vllm_ascend/ascend_forward_context.py
View File

@@ -14,7 +14,8 @@ import vllm_ascend.envs as envs_ascend
from vllm_ascend.ascend_config import get_ascend_config
from vllm_ascend.utils import (AscendDeviceType, enable_sp, flashcomm2_enable,
get_ascend_device_type, has_layer_idx,
is_moe_model)
is_moe_model,
speculative_enable_dispatch_gmm_combine_decode)
class MoECommType(Enum):
@@ -242,7 +243,7 @@ def select_moe_comm_method(num_tokens: int,
dynamic_eplb = ascend_config.dynamic_eplb or ascend_config.expert_map_record_path
# TODO: drop the EP-size guard when dispatch_ffn_combine supports larger EP sizes
# TODO: drop dynamic_eplb guard when dispatch_gmm_combine_decode supports tensor list inputs
# TODO: add guard for dispatch_gmm_combine_decode when mtp uses float while moe uses w8a8
# TODO: drop speculative method guard when dispatch_gmm_combine_decode supports w16a16
fused_mc2_enable = envs_ascend.VLLM_ASCEND_ENABLE_FUSED_MC2 and quant_type == "w8a8_dynamic" and (
not dynamic_eplb)
if num_tokens <= mc2_tokens_capacity:
@@ -250,6 +251,9 @@ def select_moe_comm_method(num_tokens: int,
if envs_ascend.VLLM_ASCEND_ENABLE_FUSED_MC2 == 1:
fused_decode_enable = fused_mc2_enable and get_ep_group(
).world_size <= 16 and (not is_draft_model)
elif envs_ascend.VLLM_ASCEND_ENABLE_FUSED_MC2 == 2:
fused_decode_enable = fused_mc2_enable and \
speculative_enable_dispatch_gmm_combine_decode(vllm_config)
moe_comm_type = MoECommType.FUSED_MC2 if fused_decode_enable else MoECommType.MC2
else:
fused_prefill_enable = fused_mc2_enable