LCLMs compress KV-caches through encoder-decoder architecture up to 1:16 more efficiently than previous methods while reducing peak memory consumption and processing time.
Encoder-decoder compressors with adaptive expansion improve KV-cache compression methods in speed and memory efficiency without significant quality loss.
Apple uses Vision-LLMs for Siri integration without requiring changes to existing apps and provides Core AI PyTorch Extensions to enable developers to run custom models on Apple hardware.
A self-learning framework for code-repair agents leverages their solution traces directly to generate targeted training tasks, achieving higher accuracy than previous approaches.
RL environments with software bugs (stale cache, reward hacks, false state transitions) generate toxic training data that sabotage agent training – systematic quality validation is necessary.
While video generation models produce visually convincing movements, visual quality does not correlate with practical executability by robots — an evaluation criterion overlooked by standard metrics.
Google releases Gemma 4 12B as an Apache-2.0-licensed multimodal model with unified architecture that runs locally on laptops with 16 GB VRAM and combines text, image, audio, and reasoning.
Hidden-state alignment reduces sampling variance, closes the student-teacher gap more effectively, and trains with less memory and computational time than output-only distillation.
