TAID-VLM-2B

TAID-VLM transfers knowledge from InternVL2-8B to an InternVL2-2B student using Sakana's distillation method.

Research release

The release targets visual instruction tasks

The card identifies English and the Mantis-Instruct dataset. Both its base architecture and teacher come from OpenGVLab; Sakana supplies the adaptation.TAID-VLM-2B official model card

Working with this release

Practical guidance. Use the image preprocessing expected by the checkpoint. Compare visual reading errors separately from errors in the language response.

Specifications & access

TAID-VLM-2B specifications
DetailValue and conditions
StatusResearch release
AccessDownloadable
Model identifiersSakanaAI/TAID-VLM-2B
Student ancestryOpenGVLab InternVL2-2BTAID-VLM-2B official model card
TeacherOpenGVLab InternVL2-8BTAID-VLM-2B official model card
TermsMIT; the card separately describes a research prototype not intended for commercial or critical deployment.TAID-VLM-2B official model card

Related models

  • An English language model distilled from a larger Qwen2 teacher.

    Launch date not confirmedResearch releaseDownloadable