{"slug": "google-research-introduces-retrieve-for-train-r4t-an-rl-compiled-diffusion-for", "title": "Google Research Introduces Retrieve-for-Train (R4T): An RL-Compiled Diffusion Retriever for 12× to 20× Faster Query Fan-Out", "summary": "Google Research introduced Retrieve-for-Train (R4T), a search framework that trains a fan-out language model with reinforcement learning once, then uses that model to synthesize training data for a 53.9M-parameter diffusion retriever that generates all retrieval directions in a single pass, running 12× to 20× faster than autoregressive fan-out. The fan-out language model is trained with groundedness, diversity, and alignment rewards to return coherent, diverse result sets. No code or model weights have been released yet.", "body_md": "Google Research has introduced Retrieve-for-Train (R4T), a framework for search that returns coherent, diverse result sets. It trains a fan-out language model with RL once, using groundedness, diversity, and alignment rewards. That model then synthesizes training data for a 53.9M-parameter diffusion retriever. The retriever generates all retrieval directions in a single pass, running 12× to 20× faster than autoregressive fan-out. No code or model weights have been released yet.\n\nThe post [Google Research Introduces Retrieve-for-Train (R4T): An RL-Compiled Diffusion Retriever for 12× to 20× Faster Query Fan-Out](https://www.marktechpost.com/2026/09/16/google-research-introduces-retrieve-for-train-r4t-an-rl-compiled-diffusion-retriever-for-12x-to-20x-faster-query-fan-out/) appeared first on [MarkTechPost](https://www.marktechpost.com).", "url": "https://wpnews.pro/news/google-research-introduces-retrieve-for-train-r4t-an-rl-compiled-diffusion-for", "canonical_source": "https://www.marktechpost.com/2026/09/16/google-research-introduces-retrieve-for-train-r4t-an-rl-compiled-diffusion-retriever-for-12x-to-20x-faster-query-fan-out/", "published_at": "2026-09-17 06:19:38+00:00", "updated_at": "2026-09-17 06:23:11.228555+00:00", "lang": "en", "topics": ["ai-research", "machine-learning", "large-language-models", "ai-search"], "entities": ["Google Research", "Retrieve-for-Train", "R4T", "MarkTechPost"], "alternates": {"html": "https://wpnews.pro/news/google-research-introduces-retrieve-for-train-r4t-an-rl-compiled-diffusion-for", "markdown": "https://wpnews.pro/news/google-research-introduces-retrieve-for-train-r4t-an-rl-compiled-diffusion-for.md", "text": "https://wpnews.pro/news/google-research-introduces-retrieve-for-train-r4t-an-rl-compiled-diffusion-for.txt", "jsonld": "https://wpnews.pro/news/google-research-introduces-retrieve-for-train-r4t-an-rl-compiled-diffusion-for.jsonld"}}