All Tools
M
OtherFreeOpen Source
MOLMO2
Open video VLMs with pointing and tracking
Apache-2.0
ABOUT
Molmo covered still images, but open video VLMs rarely point or track objects across frames. Molmo2 adds multi-image and video grounding with public 4B and 8B checkpoints so labs can train and serve pointing, tracking, and long-clip QA without closed APIs.
INTEGRATION GUIDE
1. Run Molmo2-8B in Transformers for image QA, video QA, and multi-image pointing
2. Serve Molmo2 with vLLM 0.15 plus molmo-utils for Gradio computer-use demos
3. Train Molmo2 from the open pretrain, SFT, and long-context SFT checkpoints
4. Track objects across video frames with MolmoPoint on the Molmo2 codebase
TAGS
pythonmultimodalvideovision-languageallenaipointingvllm