OFASys: Enabling Multitask Learning with One Line of Code!
Qwen
Researchers released OFASys, a framework for training multimodal multitask learning models that simplifies setup by allowing users to define tasks with a single line of code through an Instruction template. The team trained OFA+, a generalist model handling text, image, speech, video, and motion data across 23 tasks spanning 7 modalities while preserving 95% of specialist model performance. The framework abstracts away implementation complexities like data preprocessing and model configuration, enabling researchers to focus on task and modality design.
Why it matters
Intro Generalist Models are hot! We all see an opportunity towards a real generalist model by multimodal multitask learning. We previously release an opensourced unified multimodal pretrained model OFA for this goal. However, we actually met a lot of difficulties in our implementation. For example, it is hard to set up multiple tasks concerning multiple modalities, and it is hard to organize multitask learning, e.g., how to batchify your data and how to make your training stable.
