Temporal Expert Caching (TEC) for MoE inference — run MoE models that overflow GPU VRAM up to 6x faster on consumer hardware.
fitz-tec added to PyPI
The PyPI repository added a new package named 'fitz-tec' which provides Temporal Expert Caching (TEC) for MoE model inference, potentially exposing users to supply chain risks through malicious dependencies. Users installing this package could unknowingly execute harmful code or compromise their AI/ML environments. The PyPI ecosystem's lack of strict vetting for AI/ML-specific packages increases the scale of potential impact across AI developers and research institutions.