Baidu Open-Sources Wenxin Unlimited OCR 2 with 96% KV Cache Reduction

1 reports archived1 independent sourcesupdated 10/27/2026, 16:35:33

Synthesis & Latest Updates

1 Sources Cross-Validated

Baidu has released and open-sourced Wenxin Unlimited OCR 2, which reduces peak KV cache usage from 47GB to 1.77GB for hundred-page documents by introducing Sparse Reference Sliding Window Attention (SR-SWA). The model enables continuous processing of ultra-long documents without context limits, significantly lowering memory requirements and computational overhead.

LATEST/Baidu has released and open-sourced Wenxin Unlimited OCR 2, which reduces peak KV cache usage from 47GB to 1.77GB for hundred-page documents by introducing Sparse Reference Sliding Window Attention (SR-SWA). The model enables continuous processing of ultra-long documents without context limits, significantly lowering memory requirements and computational overhead.

Hourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first

量子位官网T2Oct 27, 16:35·85 pts
Baidu Open-Sources Wenxin Unlimited OCR 2 with 96% KV Cache Reduction
- Wenxin Unlimited OCR 2 uses SR-SWA to reduce KV cache usage by ~96% (from 47GB to 1.77GB).
- Supports continuous parsing of ultra-long documents (e.g., full Harry Potter series), breaking traditional context length limits.
- Officially open-sourced on GitHub and Hugging Face for immediate community deployment.
View Related Reports & Analysis Original Source