Commits · 1bbb2f9458f228f1a86f73f9eafc5681748186c0 · OpenDAS / vllm_cscc

11 Feb, 2026 3 commits
- Merge branch 'v0.11.0-dev_yql_2.11' into 'v0.11.0-dev' · 1bbb2f94
  zhuwenwen authored Feb 11, 2026
```
修复CompressedTensors的w8a16的支持,新增awq_marlin gemm的qwen72B的支持

See merge request dcutoolkit/deeplearing/vllm!429
```
  1bbb2f94
- 修复CompressedTensors的w8a16的支持 · 3d4721f2
  yangql1 authored Feb 11, 2026
  
  3d4721f2
- update Q/K/V_SCALE_CONSTANT · a4d28758
  zhuwenwen authored Feb 11, 2026
  
  a4d28758
10 Feb, 2026 6 commits
- Merge branch 'v0.11.0-dev-tmp' into 'v0.11.0-dev' · 04343d9d
  zhuwenwen authored Feb 10, 2026
```
The gfx928 architecture forces the use of the Triton gemm.

See merge request dcutoolkit/deeplearing/vllm!428
```
  04343d9d
- The gfx928 architecture forces the use of the Triton gemm. · f551bd1d
  wanglong3 authored Feb 10, 2026
  
  f551bd1d
- Merge branch 'v0.11.0-dev' of http://10.16.6.30/dcutoolkit/deeplearing/vllm into v0.11.0-dev · 6de849de
  zhuwenwen authored Feb 10, 2026
  
  6de849de
- for qwen3 dense,set VLLM_USE_OPT_RESHAPE_AND_CACHE=1 · 706e7a4f
  zhuwenwen authored Feb 10, 2026
  
  706e7a4f
- use reshape_and_cache_cuda for bf16 · eb8b6889
  zhuwenwen authored Feb 10, 2026
  
  eb8b6889
- 修awq的tp2的bug问题 · 3249f6cc
  yangql1 authored Feb 10, 2026
  
  3249f6cc
09 Feb, 2026 4 commits
- 新增awq_marlin gemm的qwen72B的支持 · 14d195dc
  yangql1 authored Feb 09, 2026
  
  14d195dc
- release rc3 · a27f634a
  zhuwenwen authored Feb 09, 2026
  
  a27f634a
- Merge branch 'v0.11.0-dev' of http://10.16.6.30/dcutoolkit/deeplearing/vllm into v0.11.0-dev · ee89bfaf
  zhuwenwen authored Feb 09, 2026
  
  ee89bfaf
- pd separation uses default scheduling · 12cdfda7
  zhuwenwen authored Feb 09, 2026
  
  12cdfda7
06 Feb, 2026 2 commits
- Merge branch 'v0.11.0-dev-guanyu-qwen2-vl-7b' into 'v0.11.0-dev' · 558d4c61
  zhuwenwen authored Feb 06, 2026
```
修复qwen2-vl-7b图像处理部分张量创建在cpu上导致报错问题

See merge request dcutoolkit/deeplearing/vllm!408
```
  558d4c61
- 修复qwen2-vl-7b图像处理部分张量创建在cpu上导致报错问题 · 185ccc2d
  guanyu1 authored Feb 06, 2026
  
  185ccc2d
03 Feb, 2026 2 commits
- Merge branch 'v0.11.0-dev-lxh-channelwise' into 'v0.11.0-dev' · 70506d98
  zhuwenwen authored Feb 03, 2026
```
V0.11.0 dev lxh channelwise

See merge request dcutoolkit/deeplearing/vllm!401
```
  70506d98
- V0.11.0 dev lxh channelwise · 9b42963d
  wanglong3 authored Feb 03, 2026
  
  9b42963d
29 Jan, 2026 1 commit
- add prepare_so_files to prepare so · c5980399
  zhuwenwen authored Jan 29, 2026
  
  c5980399
28 Jan, 2026 3 commits

Merge branch 'v0.11.0-dev_tc_opt' into 'v0.11.0-dev' · 0bf89b0c

zhuwenwen authored Jan 28, 2026

feat(moe)：新增 VLLM_USE_MOE_W16A16_TRTION 强制 Triton MoE

See merge request dcutoolkit/deeplearing/vllm!397

0bf89b0c

feat(moe)：新增 VLLM_USE_MOE_W16A16_TRTION 强制 Triton MoE · 57eb1192

laibao authored Jan 28, 2026

增加环境变量开关，禁用 Marlin W16A16 MoE 路径
强制 Triton 且权重已是 Marlin packed 时给出明确报错
Marlin 支持探测改为 best-effort（不再依赖 VLLM_USE_LIGHTOP）

57eb1192

release rc2 · 0946f6c9
zhuwenwen authored Jan 28, 2026

0946f6c9

27 Jan, 2026 5 commits
- Merge branch 'v0.11.0-dev-Q' into 'v0.11.0-dev' · a59531f8
  zhuwenwen authored Jan 27, 2026
```
V0.11.0 dev q

See merge request dcutoolkit/deeplearing/vllm!392
```
  a59531f8
- Q支持fp8 · 1fb40bd3
  xiabo authored Jan 27, 2026
  
  1fb40bd3
- Q支持fp8 · 1a3ed91b
  xiabo authored Jan 27, 2026
  
  1a3ed91b
- Merge branch 'v0.11.0-dev_fth' into 'v0.11.0-dev' · 0289bb5b
  zhuwenwen authored Jan 27, 2026
```
fused_moe_fp8接入lmslim

See merge request dcutoolkit/deeplearing/vllm!389
```
  0289bb5b
- fused_moe_fp8接入lmslim · cf40e82a
  SAC_fanth authored Jan 27, 2026
  
  cf40e82a
26 Jan, 2026 4 commits
- Merge branch 'v0.11.0-dev' into 'v0.11.0-dev' · 5f5ecd20
  zhuwenwen authored Jan 26, 2026
```
[MiniMax-M2.1] 新增minimax-m2.1工具调用

See merge request dcutoolkit/deeplearing/vllm!387
```
  5f5ecd20
- Update MiniMaxM2ReasoningParser · 54f62b3c
  chenych authored Jan 26, 2026
  
  54f62b3c
- Add minimax-m2.1 tool call and resoning parase in v0.11.0 · 44181448
  chenych authored Jan 26, 2026
  
  44181448
- update version of rc2 test · 8348926e
  zhuwenwen authored Jan 26, 2026
  
  8348926e
23 Jan, 2026 4 commits
- fix qwen3-next nn layout · 6cd43ae5
  zhuwenwen authored Jan 23, 2026
  
  6cd43ae5
- add fp8 support on bw · 88411543
  zhuwenwen authored Jan 23, 2026
  
  88411543
- only support fp8 e4m3 on nmz · 625b0b5e
  zhuwenwen authored Jan 23, 2026
  
  625b0b5e
- Merge branch 'v0.11.0-dev-wm-0122' into 'v0.11.0-dev' · a3488ab0
  zhuwenwen authored Jan 23, 2026
```
[fix]解决dp+graph+naive all2all dispatch时的维度不匹配问题

See merge request dcutoolkit/deeplearing/vllm!384
```
  a3488ab0
22 Jan, 2026 1 commit
- [fix]解决dp+graph+naive all2all dispatch时的维度不匹配问题 · d2c20c55
  王敏 authored Jan 22, 2026
  
  d2c20c55
21 Jan, 2026 5 commits
- perf(qwen3): 融合 q/k RMSNorm + RoPE · 3cd3d1e6
  zhuwenwen authored Jan 21, 2026
```
新增 VLLM_USE_FUSED_RMS_ROPE 分支，走 fused 路径
注册 torch.ops.vllm.rms_rotary_embedding_fuse（direct_register_custom_op）
cos_sin_cache 自动转 device/dtype 并缓存，避免每次重复拷贝
```
  3cd3d1e6
- update VLLM_USE_FUSED_RMS_ROPE=0 (default) · 80a6b121
  zhuwenwen authored Jan 21, 2026
```
for qwen3, VLLM_USE_FUSED_RMS_ROPE=1 (default)
```
  80a6b121
- Merge branch '011_blaslt_fp8_channelwise' into 'v0.11.0-dev' · ba73bd6f
  zhuwenwen authored Jan 21, 2026
```
feat: Supprot fp8 channle-wise matmul.

See merge request dcutoolkit/deeplearing/vllm!380
```
  ba73bd6f
- feat: Supprot fp8 channle-wise matmul. · 94c8a620
  wanglong3 authored Jan 21, 2026
  
  94c8a620
- Merge branch 'v0.11.0-dev-kvscale' into 'v0.11.0-dev' · 4dcfd0ae
  zhuwenwen authored Jan 21, 2026
```
V0.11.0 dev kvscale

See merge request dcutoolkit/deeplearing/vllm!378
```
  4dcfd0ae