method · first seen 4 Nov 2026
NUMA-aware tensor parallelism
Also: NUMA-TP
Sharding model weights across CPU NUMA sockets so each socket computes on locally resident memory.
Mentions per week
PapersCodeDiscussion
Earliest sources
- NUMA-Torch: Compiler-Driven NUMA-Aware LLM Prefill on Multi-Socket CPUsPapers · 4 Nov 2026 · Thisara Rathnayaka Weerakoon, Ishan Dewpura, Niman Ransindu, Shavin Anjitha, Tha…
Latest sources
- NUMA-Torch: Compiler-Driven NUMA-Aware LLM Prefill on Multi-Socket CPUsPapers · 4 Nov 2026 · Thisara Rathnayaka Weerakoon, Ishan Dewpura, Niman Ransindu, Shavin Anjitha, Tha…