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

[Refact]Refact MLA/SFA weight prefetch to consist with moe weight prefetch (#6629)

Browse Source 
### What this PR does / why we need it?
1. [Refact] Refact MLA/SFA weight prefetch to consist with moe weight
prefetch
2. Remove duplicated o_proj weight prefetch in forward for MLA/SFA

### Does this PR introduce _any_ user-facing change?
NA

### How was this patch tested?

1) Performance result:
Perf test data:
*) MLA:

| | 1st test | 2nd test | Output Token Throughput(Avg) | Performance
improvement percentage |
| --- | --- | --- | --- | --- |
| o_proj duplicate prefetch | 11.9669 token/s | 12.0287 token/s |
11.9978 |
| o_proj no duplicate prefetch | 12.5594 token/s | 12.6216 token/s |
12.5905 | 4.94%| |

single layer performace improve: 5%~8%

*) SFA:

| | 1st test | 2nd test | Output Token Throughput(Avg) | Performance
improvement percentage |
| --- | --- | --- | --- | --- |
| o_proj duplicate prefetch | 13.0523 token/s | 13.1084 token/s |
13.08035 | |
| o_proj no duplicate prefetch | 13.9844 token/s | 14.1678 token/s |
14.0761 | 7.6% |

- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd

---------

Signed-off-by: leo-pony <nengjunma@outlook.com>
This commit is contained in:
Nengjun Ma
2026-02-10 14:14:37 +08:00
committed by GitHub
parent 2a826b5fad
commit 66b60c9440
15 changed files with 98 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm_ascend/_310p/fused_moe/experts_selector.py
View File

@@ -57,8 +57,7 @@ def select_experts(
"""
# prefetch w1_w3_proj.weight preprocess
weight_prefetch_method = get_weight_prefetch_method()
if weight_prefetch_method:
weight_prefetch_method.maybe_prefetch_moe_weight_preprocess(hidden_states, "gate_up")
weight_prefetch_method.maybe_prefetch_moe_weight_preprocess(hidden_states, "gate_up")
topk_weights, topk_ids = _native_select_experts(
hidden_states=hidden_states,
router_logits=router_logits,