Welcome to R1-V Discussions!

[chenllliang]

👋 Welcome!

We’re using Discussions as a place to connect with other members of our community. We hope that you:

• Ask questions you’re wondering about.
• Share ideas.
• Engage with other community members.
• Welcome others and are open-minded. Remember that this is a community we
  build together 💪.

[zhexjtu]

Thanks for the great work! One question for curiosity: why tuning of the smallest model takes so much memory?

[chenllliang]

Hi, I think it is related to the rollout number N=8 in the default GRPO setting, which causes a large memory cost. You can reduce N to reduce the memory cost.

[krigeta]

Hello, I am truly impressed with your work and appreciate your efforts. I have a question: Is it possible to develop models for visual tasks, such as analyzing short videos (around 5 minutes) and images, that can also perform cognitive tasks? Specifically, I am interested in models that can run efficiently on consumer-grade PCs. Ideally, these models should be optimized to run on systems with 16GB of VRAM, or even provide an option for systems with 8GB of VRAM to handle tasks they previously couldn't.