HuggingFace

OpenAI's models were used in an attack on HuggingFace's infrastructure, revealing vulnerabilities in AI systems. HuggingFace had to rely on the Chinese open-weight model GLM 5.2 for forensic analysis after US frontier models blocked their requests due to safety restrictions.

KVBoost is a new technique that enables chunk-level KV cache reuse for HuggingFace models, achieving 5-48x faster Time-To-First-Token (TTFT) performance. The project demonstrates significant improvements in inference speed by reusing key-value cache data across multiple requests.