Flan-T5 Base

Flan-T5 Base

Quick answer: Flan-T5 Base is Google's 250M-parameter instruction-tuned encoder-decoder model from the "Scaling Instruction-Finetuned Language Models" paper (Chung et al., 2022). It is open-weight (Apache 2.0) and widely used as a small, fine-tunable backbone in agent and grounding research, including as the smallest MindAct backbone in the Mind2Web benchmark.

At a Glance

Historical significance

• One of the first widely-adopted open instruction-tuned encoder-decoder models
• Small enough (250M params) to fine-tune on a single GPU, making it a common research baseline

Modern context

• Long surpassed in raw capability by modern decoder-only LLMs
• Still used as a lightweight, fully open fine-tuning backbone in academic benchmarks and ablations