Commits · 62d065ca4ab15eba9b5237258145ffb2d9b2ca49 · change / sglang

19 Nov, 2025 3 commits
- 支持w8a8 compile，注册自定义算子解决部分断图问题 · 8da47f19
  renzhc authored Nov 19, 2025
  
  8da47f19
- 增加dcu_assign_req_to_token_pool、dcu_get_last_loc、dcu_assign_extend_cache_locs、d... · b08d561e
  shangxl authored Nov 19, 2025
```
增加dcu_assign_req_to_token_pool、dcu_get_last_loc、dcu_assign_extend_cache_locs、dcu_create_flashmla_kv_indices、dcu_create_chunked_prefix_cache_kv_indices算子
```
  b08d561e
- 修复w8a8_marlin tp pp · 158efb14
  maxiao1 authored Nov 19, 2025
  
  158efb14
18 Nov, 2025 4 commits
- forward_batch_info.py 增加环境变量 · a2ffdd60
  shangxl authored Nov 18, 2025
  
  a2ffdd60
- 修改forward_batch_info.py · e4e2b478
  shangxl authored Nov 18, 2025
  
  e4e2b478
- fa3 support qwen. · 4c45697e
  shangxl authored Nov 18, 2025
  
  4c45697e
- mtp · 27815038
  niuhb authored Nov 18, 2025
  
  27815038
15 Nov, 2025 1 commit
- 适配w8a8_marlin 高吞吐模式 · eed591c9
  maxiao1 authored Nov 15, 2025
  
  eed591c9
14 Nov, 2025 3 commits
- 使用groupgemm完成高吞吐模式适配。 · 54abdab4
  yiqa authored Nov 14, 2025
  
  54abdab4
- 适配w8a8_marlin · 25796d05
  maxiao1 authored Nov 14, 2025
  
  25796d05
- 算子融合 · 3dc7dc6c
  maxiao1 authored Nov 14, 2025
  
  3dc7dc6c
13 Nov, 2025 1 commit
- 以尽量小的改动支持torch compile，补全部分接口 · c03b2816
  renzhc authored Nov 13, 2025
  
  c03b2816
12 Nov, 2025 3 commits
- update to start mtp. · a55cb8b2
  linhai1 authored Nov 12, 2025
  
  a55cb8b2
- 使用groupgemm完成低延迟模式适配 · e7a1f94c
  maxiao1 authored Nov 12, 2025
  
  e7a1f94c
- modify codes with performance issues. · f37b05b5
  linhai1 authored Nov 12, 2025
  
  f37b05b5
11 Nov, 2025 4 commits
- support fp8_e4m3 and fp8_e5m2. · 59b01a00
  linhai1 authored Nov 11, 2025
  
  59b01a00
- support fp8_e4m3 and fp8_e5m2. · 34f0ebb1
  linhai1 authored Nov 11, 2025
  
  34f0ebb1
- support fp8_e4m3. · 04ea3540
  linhai1 authored Nov 11, 2025
  
  04ea3540
- support fp8_e4m3. · 484c5433
  linhai1 authored Nov 11, 2025
  
  484c5433
10 Nov, 2025 5 commits
- V0.5.4 dev liucong · 5f643074
  maxiao1 authored Nov 10, 2025
  
  5f643074
- fix bug on flash attention use for chunkprefill and radix cache · 8d0b2f15
  lizhigong authored Nov 10, 2025
  
  8d0b2f15
- refer to flashmla to add decode backend. · cd0b5891
  linhai1 authored Nov 10, 2025
  
  cd0b5891
- 添加环境变量SGLANG_USE_LIGHTOP 控制 lightop的融合rotaty_emb和moe_gated算子，默认禁用；修复RMSNorm:forward_hip中的错误逻辑 · f453578b
  renzhc authored Nov 10, 2025
  
  f453578b
- add merge_state_v2 in sgl_kernel · d297cda2
  lizhigong authored Nov 10, 2025
  
  d297cda2
08 Nov, 2025 1 commit
- add draft_extend support for dcu_mla. · d629db06
  linhai1 authored Nov 08, 2025
  
  d629db06
07 Nov, 2025 4 commits
- Fix Bug. · 4d106b5f
  linhai1 authored Nov 07, 2025
  
  4d106b5f
- V0.5.4 dev liucong · 5073dd76
  liucong8560 authored Nov 07, 2025
  
  5073dd76
- 修复extern报错、精度问题，支持radix cache和chunk prefill · abe30c35
  lizhigong authored Nov 07, 2025
  
  abe30c35
- Fix Bug. · e36f865d
  linhai1 authored Nov 07, 2025
  
  e36f865d
04 Nov, 2025 2 commits
- V0.5.4 dev linhai · ee775772
  linhai1 authored Nov 04, 2025
  
  ee775772
- enable custom_allreduce · 0c5532b0
  maxiao1 authored Nov 04, 2025
  
  0c5532b0
03 Nov, 2025 1 commit
- 调用vllm里custom all reduce · d2fdeac2
  maxiao1 authored Nov 03, 2025
  
  d2fdeac2
30 Oct, 2025 1 commit
- 适配qwen3-next · 477fddf2
  maxiao1 authored Oct 30, 2025
  
  477fddf2
29 Oct, 2025 2 commits
- update UNBALANCED_MODEL_LOADING_TIMEOUT_S=3600 · eb4ba1c2
  maxiao1 authored Oct 29, 2025
  
  eb4ba1c2
- 适配w8a8模型 · 4b9b337b
  maxiao1 authored Oct 29, 2025
  
  4b9b337b
28 Oct, 2025 3 commits
- 增加hipprof支持、修复异步调度中的同步问题 · f6528b74
  lizhigong authored Oct 28, 2025
  
  f6528b74
- 关闭custom_allreduce保持正确性 · a5718531
  maxiao1 authored Oct 28, 2025
  
  a5718531
- 补充 bench_serving.py里tpot等指标 · c333f125
  guobj authored Oct 28, 2025
  
  c333f125
27 Oct, 2025 1 commit
- fix fused_add_rms_norm bug · f9a026ad
  maxiao authored Oct 27, 2025
  
  f9a026ad
25 Oct, 2025 1 commit
- adaptation w4a8 tp · b80ae5e9
  maxiao1 authored Oct 25, 2025
  
  b80ae5e9