Nengjun Ma 66b60c9440 [Refact]Refact MLA/SFA weight prefetch to consist with moe weight prefetch (#6629) ...

### 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>

2026-02-10 14:14:37 +08:00

..

_310p

[Refactor]refactor 310p attention impl and add ut (#6579)

2026-02-07 09:26:26 +08:00

attention

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

2026-02-10 14:14:37 +08:00

compilation

Reapply "[Refactor] Unify full-graph parameter update logic (#6041)" (#6227) (#6231)

2026-01-26 09:04:54 +08:00

core

[MM][Bugfix] Update hf_config to hf_text_config (#5319)

2026-01-06 16:41:39 +08:00

device_allocator

[Refactor] Modify the binding logic to allocate CPU cores for each NPU card (#5555)

2026-01-13 09:21:28 +08:00

distributed

[Refactor]Refactor of vllm_ascend/distributed module (#5719)

2026-01-15 08:57:40 +08:00

eplb

[main2main] upgrade vllm main 0202 (#6560)

2026-02-05 19:31:17 +08:00

fake_weight

[CI] Add unit test framework (#1201)

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

kv_connector

[P/D] Using the cache load operator to replace the index select operator. (#6295)

2026-01-30 14:27:53 +08:00

model_loader/netloader

Revert "moe_gating_top_k" (#5512)

2025-12-30 15:05:47 +08:00

ops

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

2026-02-10 14:14:37 +08:00

patch/worker/patch_common

[Patch] Remove the patch of MiniCPM (#5975)

2026-02-09 14:07:44 +08:00

quantization

[main2main] upgrade vllm main 0202 (#6560)

2026-02-05 19:31:17 +08:00

sample

[Refactor] Import global var form vllm instead of overwirte it (#5469)

2026-01-07 18:41:45 +08:00

spec_decode

[Misc] upgrade to vllm main (#6646)

2026-02-10 14:08:59 +08:00

worker

[bugfix]Fix accuracy issue in PCP/DCP with speculative decoding (#6491)

2026-02-05 10:06:14 +08:00

__init__.py

[2/4][Refactor] Refactor torchair utils (#1892)

2025-07-21 19:43:30 +08:00

base.py

[Feature]: implement the fusion of allreduce and matmul in prefill phase when tp is enabled (#1926)

2025-07-28 15:13:37 +08:00

conftest.py

[Main2Main] Upgrade vllm commit to 0105 (#5595)

2026-01-06 08:44:29 +08:00

test_ascend_config.py

[Feature]refactor the npugraph_ex config, support online-infer with static kernel (#5775)

2026-01-20 21:31:38 +08:00

test_envs.py

[Misc] Remove redundant imported envs, using envs_ascend instead (#2193)

2025-08-14 09:33:39 +08:00

test_platform.py

[Misc] Removes unnecessary graph size re-initialization (#6280)

2026-01-27 14:38:07 +08:00

test_utils.py

[CI] Update UT CANN version to 8.5.0 for main branch (#6564)

2026-02-06 10:28:42 +08:00