← all releases
Apple·Dec 2, 2025·v0.2·open source
CLaRa-7B-Instruct
CLaRa-7B-Instruct is an instruction-tuned unified Retrieval-Augmented Generation (RAG) model developed by Apple. It integrates semantic document compression at 16× and 128× ratios, enabling efficient instruction-following question answering directly from compressed document representations.
AI AgentsLLMsML
overview
CLaRa-7B-Instruct is an instruction-tuned unified Retrieval-Augmented Generation (RAG) model developed by Apple. It integrates semantic document compression at 16× and 128× ratios, enabling efficient instruction-following question answering directly from compressed document representations.
key features
- 01Instruction-tuned for enhanced question-answering performance
- 02Built-in semantic document compression at 16× and 128× ratios
- 03Unified RAG model combining retrieval and generation in a shared latent space
- 04Differentiable top-k estimator aligning document relevance with answer quality
- 05Achieves strong performance on multi-hop QA benchmarks under compression
related productsbrowse all →
AIBuildAIopen sourceAIBuildAI is an AI agent that autonomously constructs AI models. Given a specific task, it initiates an agent loop to analyze the problem, design models, and execute training processes, all without human intervention.Step-Audio-R1open sourceStep-Audio-R1 is an advanced audio language model developed by StepFun AI, designed to enhance audio reasoning capabilities by grounding its reasoning in acoustic features. It introduces Modality-Grounded Reasoning Distillation (MGRD), an iterative training framework that shifts the model's reasoning from textual abstractions to acoustic properties, effectively addressing the 'inverted scaling' problem where performance degrades with longer reasoning. This model has demonstrated superior performance across various audio understanding and reasoning benchmarks, surpassing models like Gemini 2.5 Pro and achieving results comparable to Gemini 3 Pro.SINQopen sourceSINQ (Sinkhorn-Normalized Quantization) is a novel, fast, and high-quality quantization method designed to make any Large Language Model (LLM) smaller while preserving accuracy. It offers a plug-and-play, model-agnostic technique that delivers state-of-the-art performance for LLMs without sacrificing accuracy.