Research paper proposing method to selectively erase or steer key-value cache in transformer models for efficiency. Addresses computational bottleneck in long-context LLM inference.
Research paper on geometric approaches to robot policy learning; also appears on HuggingFace with 78 upvotes. Addresses action representation in embodied AI systems.
DeepSeek's DeepEP framework with 9,736 GitHub stars, recently updated June 2026. High-engagement open-source project from major Chinese AI lab.
Technical report on Qwen-RobotWorld, unifying embodied world modeling through language-conditioned video generation. Alibaba's approach to multimodal robot understanding.
Alibaba's Qwen-Agent framework with 16,558 GitHub stars, recently updated June 2026. Major multi-agent orchestration platform with sustained community engagement.