{"slug": "clara-bridging-retrieval-and-generation-with-continuous-latent-reasoning", "title": "CLaRa: Bridging Retrieval and Generation with Continuous Latent Reasoning", "summary": "Researchers introduced CLaRa (Continuous Latent Reasoning), a unified framework that bridges retrieval and generation in large language models by performing embedding-based compression and joint optimization in a shared continuous space. The method achieves state-of-the-art compression and reranking performance across multiple QA benchmarks, outperforming text-based fine-tuned baselines even at a compression rate of 16.", "body_md": "[content type paper](/research/)published July 2026\n\nCLaRa: Bridging Retrieval and Generation with Continuous Latent Reasoning\n\nAuthorsJie He†**, Richard He Bai, Sinead Williamson, Jeff Z. Pan†, Navdeep Jaitly, Yizhe Zhang\n\nCLaRa: Bridging Retrieval and Generation with Continuous Latent Reasoning\n\nAuthorsJie He†**, Richard He Bai, Sinead Williamson, Jeff Z. Pan†, Navdeep Jaitly, Yizhe Zhang\n\nRetrieval-augmented generation (RAG) enhances large language models (LLMs) with external knowledge but still suffers from long contexts and disjoint retrieval–generation optimization. In this work, we propose CLaRa (Continuous Latent Reasoning), a unified framework that performs embedding-based compression and joint optimization in a shared continuous space. To obtain semantically rich and retrievable compressed vectors, thereby reducing the document length fed into the generator, we introduce SCP, a key-preserving data synthesis framework based on question-answering and paraphrase supervision. CLaRa then trains the reranker and generator end-to-end via a single language modeling loss, with gradients flowing through both modules using a differentiable top-k estimator. Theoretically, this unified optimization aligns retrieval relevance with answer quality. Experiments across multiple QA benchmarks show that CLaRa achieves state-of-the-art compression and reranking performance, even at a text compression rate of 16, outperforming text-based fine-tuned baselines.\n\nCompress and Compare: Interactively Evaluating Efficiency and Behavior Across ML Model Compression Experiments\n\nSeptember 27, 2024[research area Human-Computer Interaction](/research/?domain=Human-Computer%20Interaction), [research area Tools, Platforms, Frameworks](/research/?domain=Tools%2C%20Platforms%2C%20Frameworks)[conference IEEE Visualization](/research/?event=IEEE%20Visualization)\n\n*Equal Contributors\n\nTo deploy machine learning models on-device, practitioners use compression algorithms to shrink and speed up models while maintaining their high-quality output. A critical aspect of compression in practice is model comparison, including tracking many compression experiments, identifying subtle changes in model behavior, and negotiating complex accuracy-efficiency trade-offs. However, existing compression tools poorly support…\n\nContext Tuning for Retrieval Augmented Generation\n\nDecember 18, 2023[research area Knowledge Bases and Search](/research/?domain=Knowledge%20Bases%20and%20Search), [research area Speech and Natural Language Processing](/research/?domain=Speech%20and%20Natural%20Language%20Processing)[Workshop at EACL](/research/?event=EACL%20Workshop)\n\nThis paper was accepted at the UncertaiNLP workshop at EACL 2024.\n\nLarge language models (LLMs) have the remarkable ability to solve new tasks with just a few examples, but they need access to the right tools. Retrieval Augmented Generation (RAG) addresses this problem by retrieving a list of relevant tools for a given task. However, RAG’s tool retrieval step requires all the required information to be explicitly present in the query. This is a…", "url": "https://wpnews.pro/news/clara-bridging-retrieval-and-generation-with-continuous-latent-reasoning", "canonical_source": "https://machinelearning.apple.com/research/clara-latent-reasoning", "published_at": "2026-07-15 00:00:00+00:00", "updated_at": "2026-07-15 20:16:39.447544+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "large-language-models", "natural-language-processing", "ai-research"], "entities": ["CLaRa", "Jie He", "Richard He Bai", "Sinead Williamson", "Jeff Z. Pan", "Navdeep Jaitly", "Yizhe Zhang"], "alternates": {"html": "https://wpnews.pro/news/clara-bridging-retrieval-and-generation-with-continuous-latent-reasoning", "markdown": "https://wpnews.pro/news/clara-bridging-retrieval-and-generation-with-continuous-latent-reasoning.md", "text": "https://wpnews.pro/news/clara-bridging-retrieval-and-generation-with-continuous-latent-reasoning.txt", "jsonld": "https://wpnews.pro/news/clara-bridging-retrieval-and-generation-with-continuous-latent-reasoning.jsonld"}}