[pref] qwen3_next add triton ops : fused_sigmoid_gating_delta_rule_update (#4818)

### What this PR does / why we need it?
qwen3_next add fused_sigmoid_gating_delta_rule_update op which fused
fused_gdn_gating+fused_recurrent_gated_delta_rule

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

Signed-off-by: wangxiaoxin-sherie <wangxiaoxin7@huawei.com>
Co-authored-by: wangxiaoxin-sherie <wangxiaoxin7@huawei.com>
This commit is contained in:
XiaoxinWang
2025-12-19 16:34:11 +08:00
committed by GitHub
parent 118b0ed346
commit 0cc3fc357f
5 changed files with 539 additions and 1 deletions

View File

@@ -285,3 +285,15 @@
# Future Plan:
# Remove this patch when vLLM support these operators.
#
# ** 15. File: worker/patch_qwen3_next.py**
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# 1. `vllm.model_executor.models.qwen3_next.Qwen3NextGatedDeltaNet._forward_core`
# Why:
# triton ops fused_recurrent_gated_delta_rule and fused_gdn_gating in vLLM perform not good on NPU.
# How
# add a new fused triton ops in vLLM with ascend implementation.
# Related PR (if no, explain why):
# https://github.com/vllm-project/vllm/pull/30860
# Future Plan:
# Remove this patch when vLLM support these operators.
#