Commits · ab7fba0ecee3a40dfce04f37e6c179e4ef515c38 · change / sglang · GitLab

06 Mar, 2025 11 commits
- Fix nightly ci Gsm8k & Fix flashinfer backend kvcache quant (#4147) · ab7fba0e
  yinfan98 authored Mar 07, 2025
  
  ab7fba0e
- Fix a draft model accuracy bug in eagle; support step=1; return logprob in eagle (#4134) · bc1534ff
  Lianmin Zheng authored Mar 06, 2025
```
Co-authored-by: Sehoon Kim <kssteven418@gmail.com>
Co-authored-by: SangBin Cho <rkooo567@gmail.com>
Co-authored-by: Sehoon Kim <sehoon@x.ai>
```
  bc1534ff
- fix bench serving bug (#4135) · 3a391812
  Lzhang-hub authored Mar 06, 2025
  
  3a391812
- [Minor] make the `__init__` function of model_runner.py shorter (#4132) · 98c73d71
  Lianmin Zheng authored Mar 06, 2025
  
  98c73d71
- Split the __init__ of scheduler as smaller functions. Improve the eagle tests (#4128) · fcc2e37f
  Lianmin Zheng authored Mar 06, 2025
  
  fcc2e37f
- Add a pointer to the real KV cache pool (#4113) · aee30630
  Zhiqiang Xie authored Mar 05, 2025
  
  aee30630
- Remove prefill-only-one-req (#4117) · 286e6540
  Lianmin Zheng authored Mar 05, 2025
  
  286e6540
- [Hoxfix] Fix incomplete token_to_kv_pool refactor (#4121) · 718c391f
  Wenxuan Tan authored Mar 05, 2025
  
  718c391f
- chore: bump v0.4.3.post3 (#4114) · fc671f66
  Yineng Zhang authored Mar 05, 2025
  
  fc671f66
- Online serving benchmarks of real datasets for hierarchical KV caching (#3211) · 25482edb
  Yueyang Pan authored Mar 06, 2025
```
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
```
  25482edb
- Debug radixcache: refactor recursive helper methods (#3029) · 62b362b1
  luzengxiangcn authored Mar 06, 2025
```
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
```
  62b362b1
05 Mar, 2025 9 commits
- Add update_weights_from_disk endpoint to Engine (#4102) · 70b3c6ee
  Jhin authored Mar 05, 2025
```
Co-authored-by: zhaochenyang20 <zhaochen20@outlook.com>
```
  70b3c6ee
- Fix triton kernel illegal memory issue for eagle (#4100) · ef9d3b3c
  Ke Bao authored Mar 06, 2025
  
  ef9d3b3c
- [Revision] Add fast decode plan for flashinfer mla (#4012) · fc91d08a
  Baizhou Zhang authored Mar 05, 2025
  
  fc91d08a
- Fix the moe padding conditional logic (#4081) · 71ab0dab
  HAI authored Mar 05, 2025
  
  71ab0dab
- [Eagle] Refactor eagle speculative decoding (#3986) · d3d4d767
  Ying Sheng authored Mar 05, 2025
```
Co-authored-by: Ke Bao <ISPObaoke@163.com>
```
  d3d4d767
- ROCM: AITER BLOCK GEMM (#4075) · 5be8f1ed
  yigex authored Mar 05, 2025
  
  5be8f1ed
- [QUANT] Add GPTQModel Dynamic Quantization + `lm_head` Quantization (#3790) · 56a724eb
  Qubitium-ModelCloud authored Mar 05, 2025
```
Signed-off-by: ZX-ModelCloud <zx@modelcloud.ai>
Co-authored-by: ZX-ModelCloud <zx@modelcloud.ai>
```
  56a724eb
- example: add vlm to token in & out example (#3941) · 583d6af7
  Mick authored Mar 05, 2025
```
Co-authored-by: zhaochenyang20 <zhaochen20@outlook.com>
```
  583d6af7
- [Minor] more code cleanup (#4077) · e074d84e
  Lianmin Zheng authored Mar 04, 2025
  
  e074d84e
04 Mar, 2025 12 commits
- Simplify eagle tests and TP sync in grammar backend (#4066) · 77a3954b
  Lianmin Zheng authored Mar 04, 2025
  
  77a3954b
- Revert "Fix nightly-test CI" (#4065) · 2dd7d0c5
  Lianmin Zheng authored Mar 04, 2025
  
  2dd7d0c5
- [Feature] Add test for speculative_token_map (#4016) · 0d4e3228
  William authored Mar 04, 2025
  
  0d4e3228
- fix: support gelu_new activation function in gpt2 (#3712) · 9545bfb2
  Xiuyu Li authored Mar 04, 2025
  
  9545bfb2
- [XCCL] Use xccl for xpu backend since xccl is ready in latest PyTorch. (#3954) · 61261b39
  Chen Shengzhi authored Mar 04, 2025
  
  61261b39
- [Fix & Style] Refactor the grammar backend to reduce human errors and improve readability (#4030) · 19120f71
  DarkSharpness authored Mar 04, 2025
  
  19120f71
- Fix `debug_tensor_dump_output_folder` optional key missing (#4046) · 87f671ab
  Qubitium-ModelCloud authored Mar 04, 2025
  
  87f671ab
- ROCm: update aiter and its usage to fused moe (bloat16, fp8, fp8 block-quant) (#4053) · 51d25405
  HAI authored Mar 04, 2025
  
  51d25405
- Fix breakage problem when using custom_ar (#4052) · e0a2c963
  kk authored Mar 04, 2025
  
  e0a2c963
- Fix: #3988 using blockwise_int8 (#4023) · 12f2e6c3
  Xihuai Wang authored Mar 04, 2025
  
  12f2e6c3
- Reasoning parser (#4000) · 95575aa7
  Xihuai Wang authored Mar 04, 2025
```
Co-authored-by: Lucas Pickup <lupickup@microsoft.com>
```
  95575aa7
- Fix assert options.num_stages != 0 error in the latest ROCm build image (#4049) · 11eea69e
  kk authored Mar 04, 2025
```
Co-authored-by: wunhuang <wunhuang@amd.com>
```
  11eea69e
03 Mar, 2025 8 commits
- Share target model embed and head weights for nextn (#4033) · 9fafa62d
  Ke Bao authored Mar 04, 2025
  
  9fafa62d
- Remove outdated test utils and fix links for the doc of sampling params (#3999) · 57a404fd
  Qiaolin Yu authored Mar 03, 2025
  
  57a404fd
- Misc clean up; Remove the support of jump forward (#4032) · 935cda94
  Lianmin Zheng authored Mar 03, 2025
  
  935cda94
- remove cache configs in model definitions (#4031) · 1a8f995c
  Lianmin Zheng authored Mar 03, 2025
  
  1a8f995c
- Clean up custom allreduce (#4029) · a3ab768a
  Lianmin Zheng authored Mar 03, 2025
  
  a3ab768a
- Improve code styles (#4021) · 66301e12
  Lianmin Zheng authored Mar 03, 2025
  
  66301e12
- Support penalty in overlap mode; return logprob with chunked prefill; improve... · ac238727
  Lianmin Zheng authored Mar 03, 2025
```
Support penalty in overlap mode; return logprob with chunked prefill; improve benchmark scripts (#3988)
Co-authored-by: SangBin Cho <rkooo567@gmail.com>
Co-authored-by: dhou-xai <dhou@x.ai>
Co-authored-by: Hanming Lu <hanming_lu@berkeley.edu>
```
  ac238727
- Fix nightly-test CI (#3826) · b4d34cd3
  yinfan98 authored Mar 03, 2025
  
  b4d34cd3