NVlabs/Long-RL

• highreadme#1

  Add a concise, high-level summary to the README's opening

  Why:

  CURRENT

  # Long-RL: Scaling RL to Long Sequences
  
  [](https://arxiv.org/abs/2507.07966)
  [](https://github.com/NVlabs/Long-RL)
  [](https://huggingface.co/Efficient-Large-Model/LongVILA-R1-7B)
  [](https://www.youtube.com/watch?v=ykbblK2jiEg)
  [](https://long-rl.hanlab.ai)
  
  <div align="center">
  
  [](https://www.youtube.com/watch?v=ykbblK2jiEg)
  
  </div>
  
  **Scaling RL to Long Videos [Paper]** <br />
  Yukang Chen *, Wei Huang *, Baifeng Shi, Qinghao Hu, Hanrong Ye, Ligeng Zhu, Zhijian Liu, Pavlo Molchanov, Jan Kautz, Xiaojuan Qi, Sifei Liu,Hongxu Yin, Yao Lu, Song Han <br />
  
  We introduce a full-stack framework that scales up reasoning in vision-language models (VLMs) to long videos, leveraging reinforcement learning.

  COPY-PASTE FIX

  Copy

  # Long-RL: Scaling RL to Long Sequences
  
  This repository introduces a full-stack framework for scaling reinforcement learning to enable reasoning in vision-language models (VLMs) over very long video sequences.
  
  [](https://arxiv.org/abs/2507.07966)
  [](https://github.com/NVlabs/Long-RL)
  [](https://huggingface.co/Efficient-Large-Model/LongVILA-R1-7B)
  [](https://www.youtube.com/watch?v=ykbblK2jiEg)
  [](https://long-rl.hanlab.ai)
  
  <div align="center">
  
  [](https://www.youtube.com/watch?v=ykbblK2jiEg)
  
  </div>
  
  **Scaling RL to Long Videos [Paper]** <br />
  Yukang Chen *, Wei Huang *, Baifeng Shi, Qinghao Hu, Hanrong Ye, Ligeng Zhu, Zhijian Liu, Pavlo Molchanov, Jan Kautz, Xiaojuan Qi, Sifei Liu,Hongxu Yin, Yao Lu, Song Han <br />
  
  We introduce a full-stack framework that scales up reasoning in vision-language models (VLMs) to long videos, leveraging reinforcement learning.

  OpenDoneDismiss
• mediumabout#2

  Add the project homepage URL to the About section

  Why:

  COPY-PASTE FIX

  Copy

  https://long-rl.hanlab.ai

  OpenDoneDismiss
• mediumtopics#3

  Add more specific topics for video reasoning and VLMs

  Why:

  CURRENT

  efficient-ai, large-language-models, long-sequence, multi-modality, reinforcement-learning, sequence-parallelism

  COPY-PASTE FIX

  Copy

  video-reasoning, vision-language-models, long-video-rl

  OpenDoneDismiss

Category GEO backends resolved for this scan: google/gemini-2.5-flash, deepseek/deepseek-v4-flash

• CATEGORY QUERY

  How to apply reinforcement learning effectively for reasoning over very long video sequences?

  you: not recommended
  AI recommended (in order):

  1. Acme (deepmind/acme)
  2. Perceiver IO
  3. VideoMAE (facebookresearch/VideoMAE)
  4. RLlib
  5. Ray (ray-project/ray)
  6. Stable Baselines3 (DLR-RM/stable-baselines3)
  7. TimeSformer (facebookresearch/TimeSformer)
  8. MViT (facebookresearch/mvit)
  9. TorchRL (pytorch/rl)
  10. Dopamine (google/dopamine)
  11. Gymnasium (Farama-Foundation/Gymnasium)
  12. OpenAI Gym (openai/gym)
  13. PyTorchVideo (facebookresearch/pytorchvideo)

  AI recommended 13 alternatives but never named NVlabs/Long-RL. This is the gap to close.

  Show full AI answer
• CATEGORY QUERY

  Seeking frameworks to combine large vision-language models with RL for long-sequence multi-modal tasks.

  you: not recommended
  AI recommended (in order):

  1. Hugging Face Transformers
  2. TRL (Transformer Reinforcement Learning)
  3. Stable Baselines3
  4. RLlib
  5. Acme
  6. Gymnasium
  7. PyTorch
  8. TensorFlow
  9. Minerva
  10. DreamerV3
  11. PlaNet
  12. RoboStack
  13. ROS (Robot Operating System)
  14. MoveIt

  AI recommended 14 alternatives but never named NVlabs/Long-RL. This is the gap to close.

  Show full AI answer

• Metadata completeness

  warn

  Suggestion:

• README presence

  pass

• Compared to common alternatives in this category, what is the core differentiator of NVlabs/Long-RL?

  pass
  ✓AI named NVlabs/Long-RL explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

• If a team adopts NVlabs/Long-RL in production, what risks or prerequisites should they evaluate first?

  pass
  ✓AI named NVlabs/Long-RL explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

• In one sentence, what problem does the repo NVlabs/Long-RL solve, and who is the primary audience?

  pass
  ✓AI named NVlabs/Long-RL explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?