Commits · fc5eb9e129bd35e956d50fcf993375b2c54c8b09 · OpenDAS / vllm_cscc

05 Jan, 2026 1 commit
- feat: enable shared expert overlap. · ee19dca6
  wanglong3 authored Jan 03, 2026
  
  ee19dca6
22 Dec, 2025 1 commit

[feat]1.优化ep sequence parallel,区分主模型和mtp逻辑;2.ep sequence parallel添加cudagraph... · 62f05dde

王敏 authored Dec 22, 2025

[feat]1.优化ep sequence parallel,区分主模型和mtp逻辑;2.ep sequence parallel添加cudagraph padding到tp_size;3.修复共享专家和deepep combine overlap

62f05dde

18 Dec, 2025 1 commit
- [feat]合入基于deepep的大EP · 13130b89
  王敏 authored Dec 18, 2025
  
  13130b89
21 Nov, 2025 2 commits
- add VLLM_USE_CUDA_GRAPH_SIZES(1) to use 1-24... (not only 1 2 4 8 16) · 2b47bce9
  zhuwenwen authored Nov 21, 2025
```
set VLLM_USE_LIGHTOP_FILL_MOE_ALIGN=1, VLLM_USE_OPT_ZEROS=1 and VLLM_USE_PP_SYNC=1
```
  2b47bce9
- add VLLM_USE_CUDA_GRAPH_SIZES to use 1-18... (not only 1 2 4 8 16) · d126ce21
  zhuwenwen authored Nov 21, 2025
  
  d126ce21
13 Nov, 2025 1 commit
- restore the default settings of disable_cascade_attn · e720f7cf
  zhuwenwen authored Nov 13, 2025
  
  e720f7cf
03 Nov, 2025 1 commit
- feat: w8a8_marlin 接入，通过-q slimquant_marlin开启，优化w4a8_marlin代码 · dd64aebf
  jujl1 authored Nov 01, 2025
  
  dd64aebf
15 Oct, 2025 1 commit
- support --no-enable-chunked-prefill of v1 · 2c4b2c80
  zhuwenwen authored Oct 15, 2025
  
  2c4b2c80
13 Oct, 2025 2 commits
- remove --no-enable-chunked-prefill of v1 · bdaaf39d
  zhuwenwen authored Oct 13, 2025
  
  bdaaf39d
- disable cascade_attn · b7989b07
  zhuwenwen authored Oct 13, 2025
  
  b7989b07
10 Oct, 2025 1 commit
- When VLLM_USE_FLASH_ATTN_PA=0, support the use of default block_size · 3daae57c
  zhuwenwen authored Oct 10, 2025
  
  3daae57c
27 Aug, 2025 1 commit
- feat: w4a8 marlin合入主分支, 通过-q slimquant_w4a8_marlin 可以启用 · 533d4c72
  jujl1 authored Aug 27, 2025
  
  533d4c72
22 Aug, 2025 2 commits
- update cuda_graph_sizes · 15347448
  zhuwenwen authored Aug 22, 2025
  
  15347448
- support adaptive max_num_seqs and cuda_graph_sizes · 65225897
  zhuwenwen authored Aug 22, 2025
  
  65225897
19 Aug, 2025 1 commit
- update v1 fa layout · 864c718a
  zhuwenwen authored Aug 19, 2025
  
  864c718a
10 Aug, 2025 1 commit
- 更改默认的full _cuda_graph启动方式为false · 3dad13fb
  gaoqiong authored Aug 10, 2025
  
  3dad13fb
08 Aug, 2025 1 commit
- update fa full_cuda_graph support · 513f17a4
  zhuwenwen authored Aug 08, 2025
  
  513f17a4
06 Aug, 2025 1 commit
- update VLLM_FLASH_ATTN_V1 to VLLM_USE_FLASH_ATTN_PA · 88dbf92c
  zhuwenwen authored Aug 06, 2025
  
  88dbf92c
05 Aug, 2025 1 commit
- when using VLLM_FLASH_ATTN_V1, set block_size to 64 · 80a682c7
  zhuwenwen authored Aug 05, 2025
  
  80a682c7
04 Aug, 2025 1 commit
- add step3-vl tuning · 2d364c4e
  zhuwenwen authored Aug 04, 2025
  
  2d364c4e
01 Aug, 2025 1 commit
- set default block_size to 16 · 80045bf7
  zhuwenwen authored Aug 01, 2025
  
  80045bf7
31 Jul, 2025 1 commit
- [feat]支持v1 engine mtp cudagraph · fe393be8
  王敏 authored Jul 31, 2025
  
  fe393be8
29 Jul, 2025 1 commit
- GLM-4.5 Model Support · 751c492c
  zhuwenwen authored Jul 29, 2025
  
  751c492c
25 Jul, 2025 1 commit
- [Fix] MLA only supports decode-only full CUDAGraph capture. Make sure all... · b35835a1
  zhuwenwen authored Jul 25, 2025
```
[Fix] MLA only supports decode-only full CUDAGraph capture. Make sure all cudagraph capture sizes <= max_num_seq.
```
  b35835a1
22 Jul, 2025 2 commits
- update max_seq_len_to_capture to option int · d560429c
  zhuwenwen authored Jul 22, 2025
  
  d560429c
- update max_seq_len_to_capture to int · 988fc31c
  zhuwenwen authored Jul 22, 2025
  
  988fc31c
21 Jul, 2025 1 commit
- update max_seq_len_to_capture · 72273242
  zhuwenwen authored Jul 21, 2025
  
  72273242
17 Jul, 2025 2 commits
- update adpaptive block_size · 937a3ec1
  zhuwenwen authored Jul 17, 2025
  
  937a3ec1
- update default block_size · 4a84480f
  zhuwenwen authored Jul 17, 2025
  
  4a84480f
12 Jul, 2025 1 commit
- [fix]1.优化mtp代码，解决prefix-caching不兼容问题; 2.修复v1 engine从tokenizer config中读取max_model_len导致长文本输入报错 · b16f183c
  王敏 authored Jul 12, 2025
  
  b16f183c
06 Jul, 2025 1 commit
- [V1] Support any head size for FlexAttention backend (#20467) · 9fb52e52
  Cyrus Leung authored Jul 07, 2025
```
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
```
  9fb52e52
04 Jul, 2025 1 commit
- [Model][3/N] Automatic conversion of CrossEncoding model (#20168) · 2e26f915
  wang.yuqi authored Jul 04, 2025
```
Signed-off-by: wang.yuqi <noooop@126.com>
```
  2e26f915
03 Jul, 2025 4 commits
- [Model][2/N] Automatic conversion of CrossEncoding model (#19978) · 6f1229f9
  wang.yuqi authored Jul 03, 2025
```
Signed-off-by: wang.yuqi <noooop@126.com>
```
  6f1229f9
- [Quantization] Bump to use latest bitsandbytes (#20424) · 1819fbda
  Jee Jee Li authored Jul 03, 2025
```
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
```
  1819fbda
- [Core] Move multimodal placeholder from chat utils to model definition (#20355) · b024a42e
  Cyrus Leung authored Jul 03, 2025
```
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
```
  b024a42e
- update to PIECEWISE · e0d49ac2
  zhuwenwen authored Jul 03, 2025
  
  e0d49ac2
02 Jul, 2025 3 commits
- [DP] Support external DP Load Balancer mode (#19790) · 657f2f30
  Nick Hill authored Jul 02, 2025
```
Signed-off-by: Nick Hill <nhill@redhat.com>
```
  657f2f30
- [Bugfix] Fix the max_seq_len limit of 16384 for DeepSeek models (#20322) · ccbfb1d1
  WangHuaqiang authored Jul 02, 2025
```
Signed-off-by: Wang Huaqiang <huaqiang.wang@intel.com>
```
  ccbfb1d1
- [Frontend] Support configurable mm placeholder strings & flexible video... · 2e7cbf2d
  Chenheli Hua authored Jul 01, 2025
```
[Frontend] Support configurable mm placeholder strings & flexible video sampling policies via CLI flags. (#20105)
Signed-off-by: Chenheli Hua <huachenheli@outlook.com>
```
  2e7cbf2d
01 Jul, 2025 1 commit
- [Perf] Validate @config in pre-commit instead of dynamically (#20200) · c05596f1
  Lionel Villard authored Jul 01, 2025
```
Signed-off-by: Lionel Villard <villard@us.ibm.com>
```
  c05596f1