EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
1,343 Commits 3 Branches 0 Tags
49818dbbed3e597f2879f1e583e1f7c4c5cd252f
Commit Graph

5 Commits

Author SHA1 Message Date
weichen
17259cb265 [Perf] [MoE] optimize all2allv (#3738) 
### What this PR does / why we need it?
1. Replace init_routing_v2 with token_permute to optimize performance.

Note: This pr will be merged after switching ci to CANN 8.3
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
vllm bench serve bs = 48 / rr = 10000 / 2k input -> 20k output:
before:
<img width="489" height="488" alt="image"
src="https://github.com/user-attachments/assets/268a19e6-9ab2-47f0-84a1-4f6d3bc342e2"
/>
 after:
<img width="480" height="500" alt="image"
src="https://github.com/user-attachments/assets/d9b1e628-0520-42d5-8a21-b42f7cd7abc7"
/>
- vLLM version: v0.11.0
- vLLM main:
83f478bb19

Signed-off-by: Pr0Wh1teGivee <calvin_zhu0210@outlook.com>
 2025-11-13 09:38:11 +08:00
offline893
e687d6af85 [BugFix]Fix group list type of mc2. (#4047) 
### What this PR does / why we need it?
Fix accrucy problem of eplb because of PTA upgrade.

### How was this patch tested?
Main:
    baseline:
| dataset | version | metric | mode | vllm-api-general-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 87.50 |

   EPLB:

| dataset | version | metric | mode | vllm-api-general-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 87.50 |
- vLLM version: v0.11.0
- vLLM main:
83f478bb19

Signed-off-by: offline0806 <3337230449@qq.com>
Co-authored-by: offline0806 <3337230449@qq.com>
 2025-11-07 17:41:56 +08:00
realliujiaxu
bedf223771 [Perf] move quant before allgather in Allgather EP (#3420) 
### What this PR does / why we need it?
move quant before allgather in Allgather EP, rely on
https://github.com/vllm-project/vllm-ascend/pull/3334

Deepseek R1 W8A8 performance on A2 with
`HCCL_ALGO="level0:NA;level1:pipeline"`:
| Seq length | Mean TTFT (ms) main | Mean TTFT (ms)  this PR |
|----------|----------|----------|
| 4k   |  375.21  | 364.99   |
| 16k  | 1465.23   | 1421.75  |
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?


- vLLM version: v0.11.0
- vLLM main:
83f478bb19

---------

Signed-off-by: realliujiaxu <realliujiaxu@163.com>
 2025-11-04 16:49:58 +08:00
offline893
627f20ce26 [BugFix]Fix group list type of mc2. (#3864) 
### What this PR does / why we need it?
Fix the precision issue caused by the inconsistency between the group
list type used by mc2 and that of eplb.

- vLLM version: v0.11.0rc3
- vLLM main:
83f478bb19

---------

Signed-off-by: offline0806 <3337230449@qq.com>
 2025-10-30 21:39:01 +08:00
weichen
63c363d3de [Refactor] [MoE] Rename moe-related classes & files (#3646) 
### What this PR does / why we need it?
1. Rename common_fused_moe.py to fused_moe.py.
2. Rename fused_moe_prepare_and_finalize.py / FusedMoEPrepareAndFinalize
to prepare_finalize.py / PrepareAndFinalize.
3. Rename vllm_ascend/ops/moe to vllm_ascend/ops/fused_moe.
4. Move vllm_ascend/ops/fused_moe.py to
vllm_ascend/ops/fused_moe/fused_moe.py
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
e2e & ut

- vLLM version: v0.11.0rc3
- vLLM main:
17c540a993

Signed-off-by: Pr0Wh1teGivee <calvin_zhu0210@outlook.com>
 2025-10-25 11:22:03 +08:00