OFA
Unified sequence-to-sequence model covering vision, language, and cross-modal tasks.
About
OFA (One For All) is a unified multimodal pre-trained model from Alibaba's DAMO Academy, published at ICML 2022. It casts every task, from image captioning and visual grounding to text-to-image generation and text classification, into a single sequence-to-sequence learning framework over English and Chinese.
The model topped the MSCOCO image captioning leaderboard and set strong results on VQA, visual grounding, and SNLI-VE, showing that one architecture with handcrafted task instructions could span modalities without task-specific heads.
OFA was released under Apache 2.0 with checkpoints on Hugging Face and ModelScope, becoming a widely used foundation for Chinese-English multimodal research.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 86; Trending score 60/100 from researcher assessment.