Commits · ad141d07228459d32f344e2bc121296665badd10 · OpenDAS / vllm_cscc

13 Oct, 2025 9 commits
- 去掉all2all ep相关代码 · 0b467604
  王敏 authored Oct 13, 2025
  
  0b467604
- update the envs for moe_sum and moe_align · 766663e6
  zhuwenwen authored Oct 13, 2025
  
  766663e6
- update moe_sum and moe_align · 1277ff09
  zhuwenwen authored Oct 13, 2025
  
  1277ff09
- optimize the implementation of moe_sum (lightop) · 7e68a7fe
  zhuwenwen authored Oct 13, 2025
  
  7e68a7fe
- remove --no-enable-chunked-prefill of v1 · bdaaf39d
  zhuwenwen authored Oct 13, 2025
  
  bdaaf39d
- optimize the implementation of moe_sum · 06a1bee2
  zhuwenwen authored Oct 13, 2025
  
  06a1bee2
- disable cascade_attn · b7989b07
  zhuwenwen authored Oct 13, 2025
  
  b7989b07
- support dsv32 · 633f8199
  zhuwenwen authored Oct 13, 2025
  
  633f8199
- 新增fp8—e5m2 · ed3cdc81
  zhuwenwen authored Oct 13, 2025
  
  ed3cdc81
12 Oct, 2025 1 commit
- [fix]修复开启dp并且不开启ep报vmfault · f0f159a4
  王敏 authored Oct 12, 2025
  
  f0f159a4
11 Oct, 2025 3 commits
- close profile.StartTracer · a92daffa
  maxiao1 authored Oct 11, 2025
  
  a92daffa
- fix pd send async perfomance · 1707bebe
  maxiao1 authored Oct 11, 2025
  
  1707bebe
- change tbo about cudagrah · e42a922c
  maxiao1 authored Oct 11, 2025
  
  e42a922c
10 Oct, 2025 6 commits
- roll back version.py · 394648be
  lizhigong authored Oct 10, 2025
  
  394648be
- token split by token adapt to pd separation & p2p can be used async · 1a135a9d
  maxiao1 authored Oct 10, 2025
  
  1a135a9d
- keep same version · fc5bfc66
  maxiao1 authored Oct 10, 2025
  
  fc5bfc66
- pd分离_tbo · 12291212
  maxiao1 authored Oct 06, 2025
  
  12291212
- When VLLM_USE_FLASH_ATTN_PA=0, support the use of default block_size · 3daae57c
  zhuwenwen authored Oct 10, 2025
  
  3daae57c
- fix deepseek_v2.py · 2e1d8e9a
  zhuwenwen authored Oct 10, 2025
  
  2e1d8e9a
09 Oct, 2025 4 commits
- 优化mori ep · 5dcc5cb8
  王敏 authored Oct 09, 2025
  
  5dcc5cb8
- [fix]优化mori ep · 8b791547
  王敏 authored Oct 09, 2025
  
  8b791547
- Revert "Merge branch 'v0.9.2-dev-lmcache-pd' into 'v0.9.2-dev'" · 7b30ecd8
  zhuwenwen authored Oct 09, 2025
```
This reverts merge request !219
```
  7b30ecd8
- [fix]修复不开启ep报错 · 7bf7df7f
  王敏 authored Oct 09, 2025
  
  7bf7df7f
06 Oct, 2025 1 commit
- modify engied_id str · 4bb8c6af
  yangshj1 authored Oct 06, 2025
  
  4bb8c6af
05 Oct, 2025 2 commits
- use opt rope · 8c0143db
  zhuwenwen authored Oct 05, 2025
  
  8c0143db
- fix rope acc error · b7b4e639
  zhuwenwen authored Oct 05, 2025
  
  b7b4e639
30 Sep, 2025 4 commits
- fix Z100L&K100 precision error · 9e5ec696
  zhuwenwen authored Sep 30, 2025
  
  9e5ec696
- 修复部分代码 · e0ba23b5
  王敏 authored Sep 30, 2025
  
  e0ba23b5
- 修复部分代码 · 46e26bf1
  王敏 authored Sep 30, 2025
  
  46e26bf1
- [feat]优化mori计算逻辑，支持cudagraph，按照bs*ep_size截断fused_moe的输入，共享专家不tp切分，去掉最后的allreduce · d2e57a90
  王敏 authored Sep 30, 2025
  
  d2e57a90
29 Sep, 2025 4 commits
- update moe_sum interface · 00bbf0bb
  zhuwenwen authored Sep 29, 2025
  
  00bbf0bb
- fix pp error and update min_block_size · 484fcfca
  zhuwenwen authored Sep 29, 2025
  
  484fcfca
- when v1 is set to block_size 16, switch to triton implementation · e470f6a1
  zhuwenwen authored Sep 29, 2025
  
  e470f6a1
- remove redundant shared_output · 50bed026
  zhuwenwen authored Sep 29, 2025
  
  50bed026
28 Sep, 2025 1 commit
- 修复kvcache-fp8—e5m2的不能开cp的bug · b2d14ba3
  yangql authored Sep 28, 2025
  
  b2d14ba3
26 Sep, 2025 5 commits
- update cat kernel · e0fdf4e8
  zhuwenwen authored Sep 26, 2025
  
  e0fdf4e8
- off rotary_embedding_deepseek_fuse · dc675592
  zhuwenwen authored Sep 26, 2025
  
  dc675592
- feat: moe_align_block_size 更新lightop 接口,加入对ep的支持 · 0dcc2e60
  jujl1 authored Sep 26, 2025
  
  0dcc2e60
- update op · 92c6171e
  zhuwenwen authored Sep 26, 2025
  
  92c6171e
- add PD P do pp · ecd5815f
  xiabo authored Sep 26, 2025
  
  ecd5815f