Repository navigation

Implicit language-model-only mode via limit-mm-per-prompt - #22299

Merged

vllm-bot merged 73 commits into

vllm-project:mainfrom

ywang96:main

Aug 9, 2025

Merged

Implicit language-model-only mode via limit-mm-per-prompt#22299
vllm-bot merged 73 commits into
vllm-project:mainfrom
ywang96:main

Commits on Aug 6, 2025
- Test multimodal limit disables model
  Roger Wang
  committed
  16e2d78
- update
  Roger Wang
  committed
  d666559
- add models
  Roger Wang
  committed
  7714cfa
- add llava
  Roger Wang
  committed
  a2f4720
- Merge pull request #3 from ywang96/codex/add-text-only-mode-for-llama4
  Roger Wang
  authored
  2f5da46
- add qwen2.5omni
  Roger Wang
  committed
  f9d4709
- move test
  Roger Wang
  committed
  3bc743d
- update
  Roger Wang
  committed
  7478cdc
- update
  Roger Wang
  committed
  9c66f5d
- address comment
  Roger Wang
  committed
  3048ba3
- revert changes
  Roger Wang
  committed
  31f45f5
- refactor
  Roger Wang
  committed
  dc20458
- update
  Roger Wang
  committed
  cf71e91
- update message
  Roger Wang
  committed
  f3a7a3e
- cache call
  Roger Wang
  committed
  741c1b0
- update
  Roger Wang
  committed
  6443303
- update
  Roger Wang
  committed
  8881e6b
- make mypy happy
  Roger Wang
  committed
  e862c6b
- add qwen2vl
  Roger Wang
  committed
  73a1a6e
- update message
  Roger Wang
  committed
  fc41700
- fix mypy
  Roger Wang
  committed
  69526d3
- add unittest
  Roger Wang
  committed
  68b2413
- update test
  Roger Wang
  committed
  434e9b3
- add to pipeline
  Roger Wang
  committed
  7594efb
Commits on Aug 7, 2025
- comment
  Roger Wang
  committed
  6744ac1
- comments
  Roger Wang
  committed
  e1748fa
- update
  Roger Wang
  committed
  e0f1383
- delete
  Roger Wang
  committed
  af746ee
- comment
  Roger Wang
  committed
  aac7794
- update
  Roger Wang
  committed
  1c6cef8
- update test
  Roger Wang
  committed
  5af9e58
- simplify
  Roger Wang
  committed
  15850be
- update
  Roger Wang
  committed
  12966d4
- Merge branch 'main' into main
  Roger Wang
  authored
  acf0d5a
- update
  Roger Wang
  committed
  e941f33
- update
  Roger Wang
  committed
  e329b33
- move
  Roger Wang
  committed
  9b5a176
- update
  Roger Wang
  committed
  0a6a655
Commits on Aug 8, 2025
- [Misc] normalize multiprocessing Queue usage (#22371)
  
  andyxning
  authored and
  Roger Wang
  committed
  006c0c8
- [ROCm] [V1] [SpecDec] Enable Speculative Decoding on ROCm V1 Engine (#21496)
  
  tjtanaa
  authored and
  Roger Wang
  committed
  4f060b6
- [PERF] Use pybase64 to more quickly decode prompt embeddings (#22469)
  
  qthequartermasterman
  authored and
  Roger Wang
  committed
  665698c
- Add ModelOpt Qwen3 nvfp4 support (#20101)
  
  Edwardf0t1
  authored and
  Roger Wang
  committed
  a34e38f
- Support Tensorrt-LLM MoE fp4 for low-latency (#21331)
  
  authored and
  Roger Wang
  committed
  8e049be
- Fix Flashinfer CUTLASS MOE Allgather (#21963)
  
  wenscarl
  authored and
  Roger Wang
  committed
  c7d8545
- [Kernel] Add support for block FP8 on SM120 (NVIDIA 5090 and RTX PRO 6000) (#22131)
  
  0xjunhao
  authored and
  Roger Wang
  committed
  4355cfb
- [Bugfix] Fix RuntimeError: Index put requires the source and destination dtypes match (#22065)
  
  chaunceyjiang
  authored and
  Roger Wang
  committed
  f20b040
- not tie_word_embeddings for glm-4.5 and glm-4.5v (#22460)
  
  zRzRzRzRzRzRzR
  authored and
  Roger Wang
  committed
  12b4853
- Optimize MiniCPMO mask creation with vectorized implementation (#22464)
  
  skyloevil
  authored and
  Roger Wang
  committed
  85d4756
- Fix pre-commit (#22487)
  
  DarkLight1337
  authored and
  Roger Wang
  committed
  64b3a57
- [bugfix] Fix Llama3/4 issues caused by FlashInfer 0.2.10 (#22426)
  
  nvpohanh
  authored and
  Roger Wang
  committed
  ebcb01c
- [Doc] Sleep mode documentation (#22310)
  
  authored and
  Roger Wang
  committed
  42f674c
- [bench] Fix benchmark/serve.py to ignore unavailable results (#22382)
  
  lk-chen
  authored and
  Roger Wang
  committed
  7723c45
- [CI/Build] Fix multimodal tests (#22491)
  
  DarkLight1337
  authored and
  Roger Wang
  committed
  cf8aab7
- Revert "[CI/Build] Fix multimodal tests (#22491)"
  Roger Wang
  committed
  e05914e
- Revert "[bench] Fix benchmark/serve.py to ignore unavailable results (#22382)"
  Roger Wang
  committed
  244e08f
- Revert "[Doc] Sleep mode documentation (#22310)"
  Roger Wang
  committed
  9e4da59
- Revert "[bugfix] Fix Llama3/4 issues caused by FlashInfer 0.2.10 (#22426)"
  Roger Wang
  committed
  df168cb
- Revert "Fix pre-commit (#22487)"
  Roger Wang
  committed
  4b7992f
- Revert "Optimize MiniCPMO mask creation with vectorized implementation (#22464)"
  Roger Wang
  committed
  83b7ff0
- Revert "not tie_word_embeddings for glm-4.5 and glm-4.5v (#22460)"
  Roger Wang
  committed
  9c4d6e1
- Revert "[Bugfix] Fix RuntimeError: Index put requires the source and destination dtypes match (#22065)"
  Roger Wang
  committed
  ba90fb1
- Revert "[Kernel] Add support for block FP8 on SM120 (NVIDIA 5090 and RTX PRO 6000) (#22131)"
  Roger Wang
  committed
  b613e22
- Revert "Fix Flashinfer CUTLASS MOE Allgather (#21963)"
  Roger Wang
  committed
  3d64d84
- Revert "Support Tensorrt-LLM MoE fp4 for low-latency (#21331)"
  Roger Wang
  committed
  0c2e198
- Revert "Add ModelOpt Qwen3 nvfp4 support (#20101)"
  Roger Wang
  committed
  e8661ce
- Revert "[PERF] Use pybase64 to more quickly decode prompt embeddings (#22469)"
  Roger Wang
  committed
  2d6e358
- Revert "[ROCm] [V1] [SpecDec] Enable Speculative Decoding on ROCm V1 Engine (#21496)"
  Roger Wang
  committed
  03807c7
- Revert "[Misc] normalize multiprocessing Queue usage (#22371)"
  Roger Wang
  committed
  dda8e00
- Merge branch 'vllm-project:main' into main
  Roger Wang
  authored
  1cfe05a
Commits on Aug 9, 2025
- undo
  Roger Wang
  committed
  fc98f6f
- Merge branch 'vllm-project:main' into main
  Roger Wang
  authored
  ac0fa66
- delete
  Roger Wang
  committed
  33d1f48
- update
  Roger Wang
  committed
  19ba00a

Commits

Commits on Aug 6, 2025

Commits on Aug 7, 2025

Commits on Aug 8, 2025

Commits on Aug 9, 2025