Distrifusion Distributed Parallel Inference For High Resolution Mineru diffusion supports multiple prompt types for different document parsing targets. each prompt is designed for a specific output structure rather than a single generic free form response. In this paper, we propose mineru diffusion, a 2.5b parameter diffusion based framework for doc ument ocr, replacing autoregressive decoding with block level parallel diffusion decoding and confidence guided scheduling to improve efficiency and scalability.
Mineru Opendatalab released mineru diffusion, a 2.5b parameter document ocr model that replaces the standard left to right text generation used by most systems with parallel diffusion decoding. Mineru diffusion rethinks document ocr as inverse rendering using parallel diffusion decoding, boosting throughput and accuracy even under adversarial conditions. Motivated by this insight, we propose mineru diffusion, a unified diffusion based framework that replaces autoregressive sequential decoding with parallel diffusion denoising under visual conditioning. A team from shanghai artificial intelligence laboratory and peking university published mineru diffusion — a document ocr framework that abandons classical autoregressive generation in favor of diffusion based decoding.
Mineru Motivated by this insight, we propose mineru diffusion, a unified diffusion based framework that replaces autoregressive sequential decoding with parallel diffusion denoising under visual conditioning. A team from shanghai artificial intelligence laboratory and peking university published mineru diffusion — a document ocr framework that abandons classical autoregressive generation in favor of diffusion based decoding. Unlike traditional models that generate text token by token, this approach utilizes a block wise diffusion decoder to enable parallel processing within document sections. We’re on a journey to advance and democratize artificial intelligence through open source and open science. Traditional ocr systems decode documents token by token, causing latency to scale with length and early errors to cascade through entire pages. this episode explores mineru diffusion, which reframes document parsing as "inverse rendering" using masked diffusion to fill in and revise tokens in parallel rather than left to right. The primary contribution of mineru diffusion is the successful application of parallel diffusion to the document ocr task. by moving away from the "language model" style of sequential prediction, the framework addresses the core issues of latency and hallucination that plague current vlms.
Mineru Unlike traditional models that generate text token by token, this approach utilizes a block wise diffusion decoder to enable parallel processing within document sections. We’re on a journey to advance and democratize artificial intelligence through open source and open science. Traditional ocr systems decode documents token by token, causing latency to scale with length and early errors to cascade through entire pages. this episode explores mineru diffusion, which reframes document parsing as "inverse rendering" using masked diffusion to fill in and revise tokens in parallel rather than left to right. The primary contribution of mineru diffusion is the successful application of parallel diffusion to the document ocr task. by moving away from the "language model" style of sequential prediction, the framework addresses the core issues of latency and hallucination that plague current vlms.
Comments are closed.