Multimodal model

Marlin 2B

Marlin-2B is NemoStation’s open-source 2B video-language model for dense video captioning and natural-language temporal grounding.

Type
Multimodal
Weights
Open
Downloads
5k
Likes
592
Pipeline
video text to text
License
apache-2.0

Model hub

NemoStation/Marlin-2B

Repo
NemoStation/Marlin-2B
Library
transformers
Access
Gated
Technical tags and languages
en transformers safetensors qwen3_5 text-generation video multimodal video-captioning temporal-grounding qwen VLM video-text-to-text custom_code
Open on Hugging Face

Papers

3

More from NemoStation

3

About NemoStation

1

NemoStation is a video AI research and product lab focused on building small, grounded video understanding models that convert video data into structured, machine-readable information. Flagship product: Marlin-2B, a 2B-parameter video VLM fine-tuned on Qwen3.5-2B that produces dense scene/event captions with timestamps and resolves natural-language temporal queries. State-of-the-art in its weight class on CaReBench, DREAM-1K, and TimeLens-Bench. Also produces CaReBench, a video captioning benchm