Hugging Face TRL 支援視覺語言模型 (VLM) 對齊：輕鬆實現多模態 DPO 與 ORPO 訓練
Original: Vision Language Model Alignment in TRL ⚡️
Hugging Face's TRL (Transformer Reinforcement Learning) is a popular open-source library specifically designed for aligning language models…
Hugging Face 旗下的 TRL（Transformer Reinforcement Learning）套件迎來重大更新，正式支援視覺語言模型（VLM）的對齊訓練。開發者現在可以直接使用 DPOTrainer 或 ORPOTrainer 來處理包含圖像與文字的偏好資料集。此更新簡化了 LLaVA、PaliGemma 等主流多模態模型的微調流程，並支援 QLoRA 與 DeepSpeed 等顯存優化技術，大幅降低了 VLM 對齊的門檻。
Hugging Face's TRL (Transformer Reinforcement Learning) is a popular open-source library specifically designed for aligning language models (LLMs). In its latest update, TRL has officially expanded its functionality to fully support alignment training for Vision Language Models (VLMs). This means developers can now use human preference data to optimize multimodal models' performance in visual understanding, reasoning, and conversation.
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.