LGAIMar 20

Trained Persistent Memory for Frozen Decoder-Only LLMs

arXiv:2603.2232931.8h-index: 1
AI Analysis

This work addresses the challenge of adding persistent memory to frozen decoder-only LLMs, which could enhance their ability to retain information across sessions, though it is incremental as it builds on prior encoder-decoder and brain-inspired frameworks.

The paper tackled the problem of enabling persistent memory in frozen decoder-only language models, which are stateless by default, by adapting six memory injection methods to GPT-2 and training only a small adapter. The result showed that at 1x capacity, three methods with strong architectural priors achieved retained-memory scores of 7-18% and knowledge gains of 7-10, while others failed, but at 10x capacity all methods converged, indicating the gap is architectural.

Decoder-only language models are stateless: hidden representations are discarded after every forward pass and nothing persists across sessions. Jeong (2026a) showed that trained memory adapters give a frozen encoder-decoder backbone persistent latent-space memory, building on the lateral-memory framework of Jeong (2026b,c). Here we ask whether the same principle transfers to the decoder-only setting, where no cross-attention pathway exists and memory must enter through self-attention alone. We adapt six methods -- prefix, parallel cross-attention, KV extension, Hebbian memory, context-gated branch, and slot-based sparse write -- to a frozen GPT-2, training only a small adapter $θ_{mem}$. The write rule is shared; only the read injection changes from decoder cross-attention to self-attention KV prefix or parallel branch. On LoCoMo we find a striking inductive-bias dichotomy: at $1\times$ capacity, three methods with strong architectural priors -- cross-attention (M.2), Hebbian (M.4), and slot write (M.6) -- achieve retained-memory scores of $7-18\%$ and knowledge gains $ΔK$ of $7-10$, while the other three fail ($< 0.4\%$). At $10\times$ capacity all six converge, showing the gap is architectural, not fundamental. Together with the encoder-decoder results of Jeong (2026a) and the brain-inspired modules of Jeong (2026b,c), these findings establish persistent latent-space memory as a general paradigm spanning major transformer families.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes