使用 Constrained Beam Search（受約束的束搜索）引導 🤗 Transformers 的文本生成
Original: Guiding Text Generation with Constrained Beam Search in 🤗 Transformers
In natural language generation (NLG) tasks, precisely controlling a model's output has always been a major challenge. Traditional decoding…
本文介紹 Hugging Face Transformers 支援的「受約束的束搜索（Constrained Beam Search）」技術。傳統束搜索僅依賴機率，而此技術允許開發者指定必須出現在生成結果中的特定詞彙或短語。這在機器翻譯、摘要生成及特定領域的文本生成中非常實用，能有效控制輸出品質並確保關鍵資訊不遺漏。
In natural language generation (NLG) tasks, precisely controlling a model's output has always been a major challenge. Traditional decoding strategies like "beam search" can generate semantically coherent text with higher probability, but they cannot guarantee that specific keywords or phrases will appear in the output. To meet this "forced inclusion" need, Hugging Face introduced "Constrained Beam Search" functionality into its `transformers` library.
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.