Perplexity Releases Decider v1.3 Classification Models
Perplexity released Decider v1.3 and Decider Flash v1.3, decision models that handle classification more efficiently than generative language models. Available on Hugging Face under Apache license, OpenRouter, and the Perplexity Decisions API, these models return probability distributions for binary and multi-class questions, enabling applications like routing requests between model sizes, content moderation, search ranking, and data filtering without generating free-form text.
Version 1.3 improves performance through bidirectional attention (allowing tokens to incorporate context from the entire input rather than just preceding tokens), better-curated training data, and refined training procedures. A new “locate” feature extracts and highlights specific text for named entity recognition and question-answering tasks. The models underwent three-stage training: initial training across hundreds of public datasets spanning sentiment analysis, inference, and fact-checking; targeted domain fine-tuning with LoRA; and vision capability development, with Flash distilled from the larger v1.3 model.
Read the full article
Continue reading at Hacker News →This is an AI-generated summary. Read the original for the full story.