← all releases
Zyphra·Feb 5, 2026·v1.0·open source
Online Vector Quantized Attention (OVQ)
OVQ is a novel sequence mixing layer developed by Zyphra that aims to balance memory-compute costs and long-context capabilities more effectively than standard sequence mixing layers. It utilizes a sparse memory update to significantly increase memory state size while maintaining efficient training and inference characteristics.
LLMML
overview
OVQ is a novel sequence mixing layer developed by Zyphra that aims to balance memory-compute costs and long-context capabilities more effectively than standard sequence mixing layers. It utilizes a sparse memory update to significantly increase memory state size while maintaining efficient training and inference characteristics.
key features
- 01Linear compute and constant memory complexity
- 02Sparse memory update for increased memory capacity
- 03Competitive performance on long-context tasks
- 04Efficient training and inference
- 05Theoretical basis grounded in Gaussian mixture regression
related productsbrowse all →
Protenixopen sourceProtenix is an open-source, trainable PyTorch implementation of AlphaFold 3, designed for high-accuracy biomolecular structure prediction. It aims to advance accessible and extensible research tools for the computational biology community. ([github.com](https://github.com/bytedance/Protenix?utm_source=openai))SkyRL txopen sourceSkyRL tx is an open-source library that implements a backend for the Tinker API, enabling users to set up their own Tinker-like services on personal hardware. It supports end-to-end reinforcement learning (RL) and offers significantly faster sampling. The library is designed to be modular, allowing easy prototyping of new training algorithms, environments, and execution plans without compromising usability or speed.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.