ttanzhiqiang ee40d3d850 use npu_moe_gating_top_k_softmax (#1355) ...

### What this PR does / why we need it?
The optimization solution for non-deepseek select_experts is to replace
gating_topk_softmax with softmax+topk+to, which is optimized from 37us
to 14us on bf16/fp16 of qwen3-235b

- vLLM version: v0.9.2
- vLLM main:
1a4f35e2ea

---------

Signed-off-by: ttanzhiqiang <389825161@qq.com>

2025-07-11 08:55:06 +08:00

..

__init__.py

[CI] Add unit test framework (#1201)

2025-06-16 18:32:28 +08:00

test_fused_moe.py

[CI] Add unit test framework (#1201)

2025-06-16 18:32:28 +08:00

test_gating_top_k_softmax.py

use npu_moe_gating_top_k_softmax (#1355)

2025-07-11 08:55:06 +08:00

test_multi_step.py

[CI] Add unit test framework (#1201)

2025-06-16 18:32:28 +08:00

test_rotary_embedding.py

Fix the device error when using ray as vllm-acend backend (#884)

2025-06-16 21:03:16 +08:00

test_vocabparallelembedding.py

[1/N][CI] Move linting system to pre-commits hooks (#1256)

2025-07-10 14:17:15 +08:00