Commits · 0ae9ce7517f0196738db3cb6c4f52c16aa3950de · OpenDAS / vllm_cscc

21 Nov, 2025 4 commits
- feat: pp mtp加入零消耗调度，减少空泡 · 0ae9ce75
  jujl1 authored Nov 21, 2025
  
  0ae9ce75
- feat: pipeline_parallel新增pp域请求数均衡,VLLM_USE_PP_BALANCE控制，默认开启 · d8ea775f
  jujl1 authored Oct 31, 2025
  
  d8ea775f
- 根据不同场景，更新默认调度和分离调度的选择 · 21d22cbd
  zhuwenwen authored Nov 20, 2025
  
  21d22cbd
- use schedule_split_pd (dpsk mtp + fullgraph) · 155fe0d2
  zhuwenwen authored Nov 20, 2025
  
  155fe0d2
19 Nov, 2025 8 commits
- 修复pd分离开cp引起的校检问题 · 22a14b84
  zhuwenwen authored Nov 19, 2025
```
修改schedule_default和schedule_split_pd的选择
```
  22a14b84
- update m // 8 · ea2c27ab
  zhuwenwen authored Nov 19, 2025
  
  ea2c27ab
- update VLLM_USE_PD_SPLIT=0 · 90fc8f66
  zhuwenwen authored Nov 19, 2025
  
  90fc8f66
- set VLLM_USE_PD_SPLIT=0 (dpsk-v3) · d95a8fff
  zhuwenwen authored Nov 19, 2025
  
  d95a8fff
- Merge branch 'v0.9.2-dev' of http://10.16.6.30/dcutoolkit/deeplearing/vllm into v0.9.2-dev · ad74f612
  zhuwenwen authored Nov 19, 2025
  
  ad74f612
- add qwen3-30B tp2 tn moe json · 7270e6d9
  zhuwenwen authored Nov 19, 2025
  
  7270e6d9
- set VLLM_USE_LIGHTOP_FILL_MOE_ALIGN=0 · 46006aee
  zhuwenwen authored Nov 19, 2025
  
  46006aee
- Merge branch '1117_deepseek_reduce_fusenorm_quant_push_1' into 'v0.9.2-dev' · 65f4a174
  zhuwenwen authored Nov 19, 2025
```
deepseekv2-w4a8支持custom-rms-quant融合

See merge request dcutoolkit/deeplearing/vllm!259
```
  65f4a174
18 Nov, 2025 1 commit
- deepseekv2-w4a8支持custom-rms-quant融合 · 3e6729e0
  wujl5 authored Nov 18, 2025
  
  3e6729e0
17 Nov, 2025 5 commits
- update VLLM_USE_LIGHTOP_FILL_MOE_ALIGN · 813f81fb
  zhuwenwen authored Nov 17, 2025
  
  813f81fb
- set VLLM_USE_LIGHTOP_FILL_MOE_ALIN=0 for dpsk · 70f1c878
  zhuwenwen authored Nov 17, 2025
  
  70f1c878
- add VLLM_USE_LIGHTOP_FILL_MOE_ALIN · 904b87ba
  zhuwenwen authored Nov 17, 2025
  
  904b87ba
- add VLLM_USE_LIGHTOP_FILL_MOE_ALIN · cff5c2d2
  zhuwenwen authored Nov 17, 2025
  
  cff5c2d2
- update synchronize · ed53dfb0
  zhuwenwen authored Nov 17, 2025
  
  ed53dfb0
14 Nov, 2025 1 commit
- add VLLM_USE_PP_SYNC to use pp sync · a1e9e36f
  zhuwenwen authored Nov 14, 2025
  
  a1e9e36f
13 Nov, 2025 5 commits
- update qwen3 of rmsnorm · 8375370f
  zhuwenwen authored Nov 13, 2025
  
  8375370f
- Merge branch 'v0.9.2-dev-wm-1113' into 'v0.9.2-dev' · 8fc64935
  zhuwenwen authored Nov 13, 2025
```
[fix]解决DeepSeek-V3.2 mtp启动失败

See merge request dcutoolkit/deeplearing/vllm!253
```
  8fc64935
- [fix]解决moe_fused_gate编译错误，去掉mla中mtp部分的修改 · de8e0fa8
  王敏 authored Nov 13, 2025
  
  de8e0fa8
- restore the default settings of disable_cascade_attn · e720f7cf
  zhuwenwen authored Nov 13, 2025
  
  e720f7cf
- add VLLM_USE_OPT_ZEROS to replace triton_ (torch.zeros) · b364c176
  zhuwenwen authored Nov 13, 2025
  
  b364c176
12 Nov, 2025 4 commits
- revert DEFAULT_MAX_NUM_BATCHED_TOKENS = 2048 · c340a5df
  zhuwenwen authored Nov 12, 2025
  
  c340a5df
- set default_max_num_batched_tokens = 10240 · 6d25880c
  zhuwenwen authored Nov 12, 2025
  
  6d25880c
- Merge branch 'v0.9.2-dev' of http://10.16.6.30/dcutoolkit/deeplearing/vllm into v0.9.2-dev · 1dc70839
  zhuwenwen authored Nov 12, 2025
  
  1dc70839
- update qwen3_moe of layernorm and activation · a014d6a5
  zhuwenwen authored Nov 12, 2025
  
  a014d6a5
11 Nov, 2025 2 commits
- Merge branch 'v0.9.2-dev-blaslt-w8a8-GEMM' into 'v0.9.2-dev' · 33a5ce88
  zhuwenwen authored Nov 11, 2025
```
Support blaslt w8a8 GEMM op.

See merge request dcutoolkit/deeplearing/vllm!238
```
  33a5ce88
- Support blaslt w8a8 GEMM op. · 836dee3b
  wanglong3 authored Nov 11, 2025
  
  836dee3b
10 Nov, 2025 4 commits
- Merge branch 'v0.9.2-dev-wm-1110' into 'v0.9.2-dev' · 8d6b0b0a
  zhuwenwen authored Nov 10, 2025
```
[fix]修复mtp中的笔误

See merge request dcutoolkit/deeplearing/vllm!249
```
  8d6b0b0a
- Merge branch 'v0.9.2-dev-fth' into 'v0.9.2-dev' · 475128c1
  zhuwenwen authored Nov 10, 2025
```
解决w8a8 pp16开启marlin的oom问题

See merge request dcutoolkit/deeplearing/vllm!248
```
  475128c1
- [fix]修复mtp中的笔误 · d21e2db7
  王敏 authored Nov 10, 2025
  
  d21e2db7
- 解决w8a8 pp16开启marlin的oom问题 · 21f31cf6
  SAC_fanth authored Nov 10, 2025
  
  21f31cf6
09 Nov, 2025 2 commits
- Merge branch 'v0.9.2-dev-wm-1108' into 'v0.9.2-dev' · aa71cc53
  zhuwenwen authored Nov 09, 2025
```
[fix]解决开启mtp后，在极端情况碰到显存不足时，导致mla中申请的tensor数据错乱问题

See merge request dcutoolkit/deeplearing/vllm!247
```
  aa71cc53
- set DEFAULT_MAX_NUM_BATCHED_TOKENS=10240 · 98a008db
  zhuwenwen authored Nov 09, 2025
  
  98a008db
08 Nov, 2025 1 commit
- [fix]解决开启mtp后，在极端情况碰到显存不足时，导致mla中申请的tensor数据错乱问题 · 8c552f40
  王敏 authored Nov 08, 2025
  
  8c552f40
07 Nov, 2025 3 commits
- update envs.py · 2c8a16d6
  zhuwenwen authored Nov 07, 2025
  
  2c8a16d6
- remove redundant envs · d0485181
  zhuwenwen authored Nov 07, 2025
  
  d0485181
- Merge branch 'v0.9.2-dev' of http://10.16.6.30/dcutoolkit/deeplearing/vllm into v0.9.2-dev · 4fc7b5d8
  zhuwenwen authored Nov 07, 2025
  
  4fc7b5d8