TriAttention: Optimizing the KV Cache in LLMs TriAttention, a new KV cache compression technique, improves the efficiency of large language models by reducing memory use and increasing processing speed.
Muse Spark: the new AI model and its agentic limits The new Muse Spark model focuses on multimodality and multi-agent orchestration, but the issue of agentic workflows and coding in real contexts remains clear.
Generative AI and behavioural stability of models The behavioural stability of AI models is becoming a decisive factor for reliability, safety and adoption in business processes.