1 items across the graph — tagged with Depin.
NUMA-distributed weight banking for LLM inference on IBM POWER8. 147 t/s (8.8x stock). Part of the Proof of Physical AI stack.
