aiqbench analysis
RoVE introduces a parameter-free modification to Rotary Position Embeddings (RoPE) that renders the value pathway position-sensitive, effectively transforming RoPE attention into an attentive convolution mechanism. Empirical results on GPT-2 models demonstrate consistent improvements in few-shot learning and long-context retrieval by better aggregating information across distant tokens. This advancement offers a theoretically unified perspective for enhancing positional awareness without introducing additional trainable parameters to model architectures.
RoVE: Rotary Value Embeddings Attention for Relative Position-dependent Value Pathways
arXiv:2606.11275v2 Announce Type: replace-cross Abstract: Rotary Position Embeddings (RoPE) make attention scores position-relative but leave the value pathway position-blind: the message sent by a value token is the same regardless of its distance from the query. We propose RoVE, a parameter-free modification that makes values position-sensitive by rotating them simultaneously with keys, and show that it turns RoPE attention into attentive convolution. This new perspective unifies several independent formulations of the same operation across computer vision, robotics, and modern LLM architectures. Trained 124M and 354M GPT-2 models show consistent empirical gains over RoPE on few-shot in-context learning, out-of-distribution perplexity, and long-context retrieval, with the clearest improvements on tasks that require long-range aggregation.
Abstract:Rotary Position Embeddings (RoPE) make attention scores position-relative but leave the value pathway position-blind: the message sent by a value token is the same regardless of its distance from the query. We propose RoVE, a parameter-free modification that makes values position-sensitive by rotating them simultaneously with keys, and show that it turns RoPE attention into attentive convolution. This new perspective unifies several independent formulations of the same operation across computer vision, robotics, and modern LLM architectures. Trained 124M and 354M GPT-2 models show consistent empirical gains over RoPE on few-shot in-context learning, out-of-distribution perplexity, and long-context retrieval, with the clearest improvements on tasks that require long-range aggregation.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2606.11275 [cs.LG]
(or arXiv:2606.11275v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2606.11275 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 202 words
Related coverage
RoVE: Rotary Value Embeddings Attention for Relative Position-dependent Value Pathways
arXiv · Jun 12, 2026
Beyond Sequence Order: Syntax-Informed Positional Embeddings for Transformers
arXiv · Aug 7, 2026
MeRoPE: Metric Rotary Position Embedding for Camera-Controlled Video Generation
arXiv · Sep 2, 2026
Journey Operators for Structured Multi-Axis Composition
arXiv · Jul 31, 2026