A heterogeneous multi-tier memory fabric for MoE and LLM inference. Hot AI objects are promoted to executable VRAM, while cold objects are transparently tiered across heterogeneous GPU memory, DDR, PMEM/CXL and storage.

2 Open Issues Need Help Last updated: Aug 8, 2026

Open Issues Need Help

View All on GitHub

A heterogeneous multi-tier memory fabric for MoE and LLM inference. Hot AI objects are promoted to executable VRAM, while cold objects are transparently tiered across heterogeneous GPU memory, DDR, PMEM/CXL and storage.

Python

A heterogeneous multi-tier memory fabric for MoE and LLM inference. Hot AI objects are promoted to executable VRAM, while cold objects are transparently tiered across heterogeneous GPU memory, DDR, PMEM/CXL and storage.

Python