Commits · 50bed0264fbc3129c83441b727b3402bcd622243 · OpenDAS / vllm_cscc

29 Sep, 2025 2 commits
- remove redundant shared_output · 50bed026
  zhuwenwen authored Sep 29, 2025
  
  50bed026
- Merge branch 'v0.9.2-dev' into v0.9.2-dev-ds · 42b06117
  zhuwenwen authored Sep 29, 2025
  
  42b06117
28 Sep, 2025 3 commits
- 修复kvcache-fp8—e5m2的不能开cp的bug · b2d14ba3
  yangql authored Sep 28, 2025
  
  b2d14ba3
- 新增fp8—e5m2_utils · bb13d854
  yangql authored Sep 28, 2025
  
  bb13d854
- 新增fp8—e5m2 · e29550e2
  yangql authored Sep 28, 2025
  
  e29550e2
26 Sep, 2025 12 commits
- Merge branch 'v0.9.2-dev' of http://10.16.6.30/dcutoolkit/deeplearing/vllm into v0.9.2-dev · 48114bb1
  zhuwenwen authored Sep 26, 2025
  
  48114bb1
- update cat kernel · e0fdf4e8
  zhuwenwen authored Sep 26, 2025
  
  e0fdf4e8
- off rotary_embedding_deepseek_fuse · dc675592
  zhuwenwen authored Sep 26, 2025
  
  dc675592
- Merge branch 'v0.9.2-dev-ep' into 'v0.9.2-dev' · f331f103
  zhuwenwen authored Sep 26, 2025
```
feat: moe_align_block_size 更新lightop 接口,加入对ep的支持

See merge request dcutoolkit/deeplearing/vllm!216
```
  f331f103
- feat: moe_align_block_size 更新lightop 接口,加入对ep的支持 · 0dcc2e60
  jujl1 authored Sep 26, 2025
  
  0dcc2e60
- Merge branch 'v0.9.2-dev' of http://10.16.6.30/dcutoolkit/deeplearing/vllm into v0.9.2-dev · aa4bc884
  zhuwenwen authored Sep 26, 2025
  
  aa4bc884
- update op · 92c6171e
  zhuwenwen authored Sep 26, 2025
  
  92c6171e
- Merge branch 'v0.9.2-dev-pdpp2' into 'v0.9.2-dev' · 8db3c41c
  zhuwenwen authored Sep 26, 2025
```
add PD P do pp

See merge request dcutoolkit/deeplearing/vllm!215
```
  8db3c41c
- add PD P do pp · ecd5815f
  xiabo authored Sep 26, 2025
  
  ecd5815f
- Merge branch 'v0.9.2-dev_pp_mtp' into 'v0.9.2-dev' · a857453f
  zhuwenwen authored Sep 26, 2025
```
[fix] pp+mtp bs 1 correctness

See merge request dcutoolkit/deeplearing/vllm!214
```
  a857453f
- VLLM_USE_LIGHTOP and VLLM_USE_OPT_CAT · c1ece8c6
  zhuwenwen authored Sep 26, 2025
```
add shared_output and routed_scaling_factor of CompressedTensorsW8A8Int8MoEMethod
```
  c1ece8c6
- [fix] pp+mtp bs 1 correctness · 22d6f9af
  lizhigong authored Sep 26, 2025
  
  22d6f9af
25 Sep, 2025 4 commits
- Merge branch '0.9.2+das.opt1.rc2.dtk25041-ds-w8a8-rmsquant-push' into 'v0.9.2-dev' · c7d0b817
  zhuwenwen authored Sep 25, 2025
```
DeepSeek-R1-Channel-INT8调用rmsquant融合。

See merge request dcutoolkit/deeplearing/vllm!213
```
  c7d0b817
- add shared_output and routed_scaling_factor of w4a8 · b33ff2d6
  zhuwenwen authored Sep 25, 2025
  
  b33ff2d6
- DeepSeek-R1-Channel-INT8调用rmsquant融合。 · a5043e83
  wujl5 authored Sep 25, 2025
  
  a5043e83
- [kernels] add fused_rms_norm_contiguous and rotary_embedding_deepseek_fuse · 49810c37
  zhuwenwen authored Sep 25, 2025
```
[kernels] update moe_align_block_size and moe_sum interface
```
  49810c37
24 Sep, 2025 5 commits
- Merge branch 'v0.9.2-dev_pp_mtp' into 'v0.9.2-dev' · ddf9d10c
  zhuwenwen authored Sep 24, 2025
```
fix scheduler issu in pp + mtp

See merge request dcutoolkit/deeplearing/vllm!212
```
  ddf9d10c
- update VLLM_USE_OPT_CAT · 4d97c5fc
  zhuwenwen authored Sep 24, 2025
  
  4d97c5fc
- [kernel] add lightop's moe_sum(mul+add) fusion operator for deepseek · 8d2cac26
  zhuwenwen authored Sep 24, 2025
```
[FIX] 修复mtp和VLLM_USE_TRITON_CAT不能一起开的bug
```
  8d2cac26
- 修复mtp和VLLM_USE_TRITON_CAT不能一起开的bug · cc2dca96
  SAC_fanth authored Sep 24, 2025
  
  cc2dca96
- fix scheduler issu in pp + mtp · 2cb921da
  lizhigong authored Sep 24, 2025
  
  2cb921da
23 Sep, 2025 9 commits
- add convert_vertical_slash_indexes and convert_vertical_slash_indexes_mergehead · 5086453d
  zhuwenwen authored Sep 23, 2025
  
  5086453d
- Merge branch 'v0.9.2-dev' into v0.9.2-dev-ds · f872f0ad
  zhuwenwen authored Sep 23, 2025
  
  f872f0ad
- Merge remote-tracking branch 'origin/v0.9.2-dev-yql-kvfp8' into v0.9.2-dev · a0895c00
  zhuwenwen authored Sep 23, 2025
  
  a0895c00
- 修复kvcachefp8与cp/pc的冲突 · 8b7daa0d
  yangql authored Sep 23, 2025
  
  8b7daa0d
- 修复W4A8 PP8崩溃问题 · 4020670f
  zhuwenwen authored Sep 23, 2025
  
  4020670f
- Merge branch 'v0.9.2-dev' into v0.9.2-dev-ds · dbb27c0e
  zhuwenwen authored Sep 23, 2025
  
  dbb27c0e
- Update w4a8_utils.py · 7904da3f
  zhuwenwen authored Sep 23, 2025
  
  7904da3f
- Merge branch '092-dev_silu-mul-quant-push' into 'v0.9.2-dev' · 3320343d
  zhuwenwen authored Sep 23, 2025
```
deepseek-r1-w4a8 mlp/moe调用silu-mul-quant融合

See merge request dcutoolkit/deeplearing/vllm!209
```
  3320343d
- Merge branch 'v0.9.2-dev-pd' into 'v0.9.2-dev' · 440fc9db
  zhuwenwen authored Sep 23, 2025
```
修改P等待问题

See merge request dcutoolkit/deeplearing/vllm!210
```
  440fc9db
22 Sep, 2025 2 commits
- deepseek-r1-w4a8 mlp/moe调用silu-mul-quant融合 · 3964e019
  wujl5 authored Sep 22, 2025
  
  3964e019
- 修改P等待问题 · 560ffc4c
  xiabo authored Sep 22, 2025
  
  560ffc4c
20 Sep, 2025 1 commit
- 增加w4a8marlin-ds的moe融合算子输入 · 26c82c27
  SAC_fanth authored Sep 20, 2025
  
  26c82c27
18 Sep, 2025 2 commits
- update fused_experts_impl_w4a8 · 29ee4475
  zhuwenwen authored Sep 18, 2025
  
  29ee4475
- [kernel] add VLLM_USE_DEEPSEEK_MOE_SUM_MUL_AND to use lightop's moe_sum fusion... · 0975d9e8
  zhuwenwen authored Sep 18, 2025
```
[kernel] add VLLM_USE_DEEPSEEK_MOE_SUM_MUL_AND to use lightop's moe_sum fusion operator for deepseek
```
  0975d9e8