TF-Engram introduces SSD-backed train-free memory for LLMs

The authors present TF-Engram, a train-free Engram system that constructs phrase-specific semantic memory offline from external corpora and stores large memory tables across a GPU-DRAM-SSD hierarchy. It utilizes Early-Exit Guided Predictive Prefetching to hide external-memory latency during autoregressive decoding.

On Qwen3-0.6B, TF-Engram improves the average downstream score from 57.6 to 59.4, outperforming both the frozen backbone and a parameter-matched LoRA baseline.

These results demonstrate that static phrase memory can be integrated into LLM inference as a scalable, train-free, and low-overhead system component.