微調 Microsoft Florence-2：微軟頂尖視覺語言模型實戰指南
Original: Fine-tuning Florence-2 - Microsoft's Cutting-edge Vision Language Models
Microsoft open-sourced Florence-2 in June 2024 — a vision-language model (VLM) based on a sequence-to-sequence architecture. Despite its…
微軟推出的 Florence-2 是一款強大且輕量的視覺語言模型（VLM），僅有 232M 與 770M 兩種參數版本，卻能高效處理 OCR、目標檢測、圖像描述等多種任務。Hugging Face 官方部落格發布了這篇實用指南，詳細教學如何使用 Hugging Face 的 transformers 與 peft 函式庫，在自訂資料集上對 Florence-2 進行微調（Fine-tuning），並利用 LoRA 技術降低顯示記憶體需求，非常適合想在邊緣裝置或有限資源下部署視覺 AI 的開發者。
Microsoft open-sourced Florence-2 in June 2024 — a vision-language model (VLM) based on a sequence-to-sequence architecture. Despite its compact size (the Base version has only 232 million parameters, while the Large version has 770 million), it demonstrates remarkably strong performance across a wide range of vision tasks. What makes Florence-2 unique is that it unifies all vision tasks (such as image captioning, object detection, referring expression comprehension, OCR, and more) into "text generation" tasks, guided by specific task prefixes (for example, `<CAPTION>`, `<OD>`).
Free shows the 3-line summary; Pro unlocks the full deep summary (~300 words) so you never have to click through.
See Pro plans →Want the original English / full article?
Read on Hugging Face Blog →Summaries are AI-generated; the original article is authoritative.