Commits · 24523a1c297f33ded512127c990b0b7bf2251bf2 · OpenDAS / dynamo · GitLab

02 Apr, 2026 3 commits
- feat(vllm): add self-benchmark mode to InstrumentedScheduler (#7779) · 24523a1c
  Hongkuan Zhou authored Apr 01, 2026
  
  24523a1c
- fix: use dict sampling params in prefill warmup (#7797) · a873045c
  weireweire authored Apr 02, 2026
```
Co-authored-by: Weiliangl User <weiliangl@login-node.hosted.internal>
```
  a873045c
- refactor(protocols): deprecate cache control (#7790) · c09ac697
  ishandhanani authored Apr 01, 2026
  
  c09ac697
01 Apr, 2026 8 commits
- fix: Improve failure handling for profiling failures (#7728) · 546d6a1e
  Ashna Mehrotra authored Apr 01, 2026
```
Signed-off-by: ashnamehrotra <ashnamehrotra@gmail.com>
```
  546d6a1e
- fix: remove forcing the kv block size as 16 in vLLM backend (#7690) · f79b3d62
  Richard Huo authored Apr 01, 2026
  
  f79b3d62
- fix(sglang): use incremental_streaming_output instead of deprecated stream_output (#7642) · 8fe2082c
  YAMY authored Apr 01, 2026
```
Signed-off-by: Yangmin Li <yangminl@nvidia.com>
```
  8fe2082c
- test(planner): isolate planner-family suites [DYN-2534] (#7723) · ab5a31b5
  Alec authored Mar 31, 2026
  
  ab5a31b5
- feat: add least-loaded router (#6314) · cc22114d
  kornelcsernai-harmonic authored Mar 31, 2026
```
Signed-off-by: Kornel Csernai <239206175+kornelcsernai-harmonic@users.noreply.github.com>
Signed-off-by: PeaBrane <yanrpei@gmail.com>
Co-authored-by: PeaBrane <yanrpei@gmail.com>
```
  cc22114d
- feat: propagate OTEL trace context across SGLang multimodal E/P/D workers (#7580) · 8886d184
  jh-nv authored Mar 31, 2026
  
  8886d184
- fix: multimodal ImageLoader errors (#7703) · a8894467
  Qi Wang authored Mar 31, 2026
```
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
```
  a8894467
- refactor: remove dedicated multimodal PD worker (#7556) · 4b1d442b
  GuanLuo authored Mar 31, 2026
```
Signed-off-by: Guan Luo <41310872+GuanLuo@users.noreply.github.com>
```
  4b1d442b
31 Mar, 2026 6 commits
- feat: add initial audio/TTS pipeline support for vLLM-omni backend (#7495) · cbbde3d0
  Zhongdongming Dai authored Mar 31, 2026
```
Signed-off-by: Zhongdongming Dai <zhongdongmin@nvidia.com>
```
  cbbde3d0
- feat(container): add standalone dynamo-planner image [DYN-2533] (#7696) · a20b9dde
  Alec authored Mar 31, 2026
  
  a20b9dde
- fix: align video diffusion pipeline with TRT-LLM 1.3.0rc9 API changes (#7529) · 10fb23d2
  Indrajit Bhosale authored Mar 31, 2026
```
Signed-off-by: Indrajit Bhosale <iamindrajitb@gmail.com>
```
  10fb23d2
- feat: Add standardized Dynamo error types as Python exceptions (#7136) · 8ff0b6e7
  Tanmay Verma authored Mar 31, 2026
  
  8ff0b6e7
- fix(mocker): cut AIC replay overhead (#7692) · 4abafffe
  Yan Ru Pei authored Mar 30, 2026
```
Signed-off-by: PeaBrane <yanrpei@gmail.com>
```
  4abafffe
- refactor: restructure planner package into subpackage hierarchy (#7689) · 95c45509
  Hongkuan Zhou authored Mar 30, 2026
```
Signed-off-by: hongkuanz <hongkuanz@nvidia.com>
```
  95c45509
30 Mar, 2026 5 commits
- chore: bump to vllm_omni 0.18.0 (#7685) · 19447d66
  Ayush Agarwal authored Mar 30, 2026
```
Signed-off-by: ayushag <ayushag@nvidia.com>
```
  19447d66
- fix: Anthropic streaming double-parsing + reasoning_content roundtrip (#7358) · f0d3ce63
  MatejKosec authored Mar 30, 2026
```
Signed-off-by: Matej Kosec <mkosec@nvidia.com>
```
  f0d3ce63
- fix: remove omni dependency from dynamo.vllm path (#7683) · d58a6881
  Ayush Agarwal authored Mar 30, 2026
```
Signed-off-by: ayushag <ayushag@nvidia.com>
```
  d58a6881
- fix: race condition in multi modal PD worker (#7679) · f6e5023b
  Sergey Plotnikov authored Mar 30, 2026
```
Signed-off-by: Sergey Plotnikov <sergey.plotnikov@intel.com>
Co-authored-by: ishandhanani <82981111+ishandhanani@users.noreply.github.com>
```
  f6e5023b
- fix: make sleep/wake errors fatal in failover mode and correct RO memory accounting (#7681) · a565f105
  mohammedabdulwahhab authored Mar 30, 2026
```
Signed-off-by: mohammedabdulwahhab <furkhan324@berkeley.edu>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
```
  a565f105
28 Mar, 2026 1 commit
- feat: Support a dynamic default max_tokens for TensorRT-LLM backend (#5152) · 98d0ce91
  Steven Murr authored Mar 27, 2026
```
Signed-off-by: Steven Murr <stevemurr@gmail.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
```
  98d0ce91
27 Mar, 2026 1 commit
- chore(mocker): batch live output signal sends (#7647) · a2154ba5
  Yan Ru Pei authored Mar 26, 2026
```
Signed-off-by: PeaBrane <yanrpei@gmail.com>
```
  a2154ba5
26 Mar, 2026 1 commit
- refactor: load planner using new forwardpass metric and many improvements (#7351) · a58bcc31
  Hongkuan Zhou authored Mar 25, 2026
```
Signed-off-by: hongkuanz <hongkuanz@nvidia.com>
```
  a58bcc31
25 Mar, 2026 6 commits
- feat(mocker): add offline disagg replay (#7617) · 02b1c58a
  Yan Ru Pei authored Mar 25, 2026
```
Signed-off-by: PeaBrane <yanrpei@gmail.com>
```
  02b1c58a
- refactor(vllm): scope ray import to scale_elastic_ep, remove module-level ray state (#7618) · 4b8826b3
  Tzu-Ling Kan authored Mar 25, 2026
```
Signed-off-by: tzulingk <tzulingk@nvidia.com>
```
  4b8826b3
- feat(sglang): disagg DP rank routing + backwards-compatible network imports (#6736) · 7edb07b5
  ishandhanani authored Mar 25, 2026
```
Co-authored-by: huitian bai <baihuitian.bht@gmail.com>
```
  7edb07b5
- fix: strip tools from chat template when tool_choice=none (#7391) · 14a6122e
  Yifan Jiang authored Mar 24, 2026
```
Signed-off-by: Yifan Jiang <19356972+yifjiang@users.noreply.github.com>
Co-authored-by: Ayush Agarwal <ayushag@nvidia.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
```
  14a6122e
- feat: add power-of-two-choices router mode (#7614) · db49dfd2
  Aaron Batilo authored Mar 24, 2026
```
Signed-off-by: Aaron Batilo <abatilo@coreweave.com>
```
  db49dfd2
- fix: sglang eagle bigram tokens kv event report. (#6872) · 2fe37a51
  huitian bai authored Mar 25, 2026
```
Signed-off-by: baihuitian <baihuitian.bht@gmail.com>
Signed-off-by: PeaBrane <yanrpei@gmail.com>
Co-authored-by: PeaBrane <yanrpei@gmail.com>
```
  2fe37a51
24 Mar, 2026 3 commits
- feat: remove localhost fallback from NIXL side-channel host detection (#6539) · ee3de9e4
  Alec authored Mar 24, 2026
```
Signed-off-by: alec-flowers <aflowers@nvidia.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
```
  ee3de9e4
- feat(vllm): elastic EP scaling — ray.nodes() patch, scale_elastic_ep route,... · daa4fb04
  Tzu-Ling Kan authored Mar 24, 2026
```
feat(vllm): elastic EP scaling — ray.nodes() patch, scale_elastic_ep route, null-tolerant discovery (#7481)
Signed-off-by: tzulingk <tzulingk@nvidia.com>
```
  daa4fb04
- chore: remove deprecated automatic kv events config for vLLM LLM-94 (#7591) · 115512ef
  Alec authored Mar 23, 2026
```
Signed-off-by: alec-flowers <aflowers@nvidia.com>
```
  115512ef
23 Mar, 2026 6 commits
- feat(gms): e2e failover support for vLLM (#6818) · 1c65a588
  mohammedabdulwahhab authored Mar 23, 2026
```
Signed-off-by: mohammedabdulwahhab <furkhan324@berkeley.edu>
```
  1c65a588
- chore(snapshot): gms + snapshot support (#7026) · b2aefc53
  Janelle Cai authored Mar 23, 2026
```
Signed-off-by: Janelle Cai <jcai18@mit.edu>
```
  b2aefc53
- fix: ProphetPredictor next-step timestamp missing step_size fa… (#7208) · 4159c6fd
  liuhuijiayou authored Mar 24, 2026
```
Signed-off-by: liuhui <liuhui06@kuaishou.com>
Co-authored-by: liuhui <liuhui06@kuaishou.com>
```
  4159c6fd
- fix: vllm forwardpass metric for async scheduling (#7537) · cbdab502
  Hongkuan Zhou authored Mar 23, 2026
```
Signed-off-by: hongkuanz <hongkuanz@nvidia.com>
```
  cbdab502
- feat(mocker): add multi-worker replay and router startup fixes (#7553) · b7fe46b1
  Yan Ru Pei authored Mar 23, 2026
```
Signed-off-by: PeaBrane <yanrpei@gmail.com>
```
  b7fe46b1
- fix(snapshot): persist CUDA namespace PIDs in manifest (#7539) · ba9a8a9f
  Schwinn Saereesitthipitak authored Mar 23, 2026
  
  ba9a8a9f