QVal paper addresses evaluation of dense supervision signals for long-horizon LLM agent training. Provides methodology for assessing agent learning effectiveness.
Huawei released OpenPangu-2.0-Flash, a 92B parameter model with mixture-of-experts achieving 6B active parameters per inference. Major open-source model release.
VibeVoice 1.5B released via audio.cpp achieving 4.08x real-time performance on long-form audio (90-min podcast in 22.95 min). Native C++/ggml implementation.
Meta developed custom CXL 2.0 chip enabling DDR4 server memory reuse in DDR5-only infrastructure. Addresses hardware cost inflation for AI infrastructure.
Research paper proposes system-level architecture separating instruction and data channels in LLM agents to mitigate prompt injection attacks.