We present MoNe, a lightweight modular neural memory that attaches to any frozen pretrained Transformer to enable long-context inference without retraining. MoNe reads context in fixed-size segments via test-time learning of fast-weight neural memory networks with layer-localized gradient updates; at inference, the memory generates keys and values from the query tokens alone, with no context tokens re-read. This two-phase design decouples inference cost from context length, achieving O(N) preprocessing and O(1) query cost with peak GPU memory that does not grow with N. At 128K tokens, MoNe reduces both compute and peak GPU memory by approximately 80% compared to ICL with only 6.4% parameter overhead. MoNe generalizes to context lengths far beyond the backbone's native window, achieving strong performance on needle-in-a-haystack and word extraction benchmarks from RULER, where ICL degrades sharply.
Figures & tables
Figure 1: Comparison of ICL and MoNe. ( Left ) In ICL, question tokens must attend to all context tokens. ( Right ) In MoNe, segments are processed sequentially during test-time learning to update neural memory; at inference, question tokens attend only to memory tokens.
Figure 2: End-to-end peak GPU memory and total FLOPs for MoNe (test-time learning + inference) at context lengths of 32K and 128K, in comparison with ICL.
Figure 3: An overview of the update process of MoNe during test-time learning. For each layer, neural memory fast weights are updated using the associative memory loss, which is localized and computed entirely from layer l ’s own forward-pass activations, making the memory update both efficient and truly modular.
Figure 4: Ablation study on performance-efficiency trade-off: ( Left ) Layer selection. Attaching MoNe to all 24 decoder layers achieves the highest accuracy across all context lengths; restricting to the last 16 layers (last 32 ) yields modest degradation beyond 32K, while using only the last 8 layers (last 31 ) collapses beyond 32K. ( Right ) Segment size. T=512 (default) generalizes from 4K training contexts to 128K evaluation contexts, whereas T=128 and T=256 degrade rapidly past 16K, yet offer only marginal efficiency gains over the default.
Transformers are efficient to pre-train at scale because attention permits parallel computation across sequence positions, but long-context inference requires increasingly large KV caches. Efficient recurrent alternatives reduce the asymptotic complexity of inference, but their sequential dependence limits token-wise parallelism during training, making them harder to scale. Further, the current paradigm of extending long context tasks creates competing notions of retaining information precisely and keeping long-context inference efficient. To address this gap, we introduce Titan-Llama, a post-training method for retrofitting attention-based language models with a hierarchical memory architecture that updates parameters at test-time. Titan-Llama applies a blockwise attention mask and preserves full-fidelity attention over recent tokens while compressing historical context into recurrent, fixed-size neural memory modules (NMMs). Our formulation allows pre-trained attention layers to be replaced without training a new language model from scratch. On Llama-3.2-1B-Instruct, Titan-Llama raises accuracy over segmented attention windows from 23.87% to 40.51% on QuALITY and from 14.50% to 40.75% on LongHealth. Inference benchmarks show more than 20% faster prefill, 37% faster decode at medium context lengths, and 24% lower steady-state decode memory usage in the evaluated configurations. Together, these results demonstrate that incorporating memory architecture mechanisms that utilize test-time training effectively modulates long-context behavior of pre-trained models, offering a practical route to adapting pre-trained transformers for efficient long-context inference. Our code is released on Github GitHub
Sarah Pan, Ali Behrouz
Massachusetts Institute of Technology · Cornell University
Efficient long-context inference is not only about reducing memory cost, but also about keeping useful contextual evidence accessible as generation proceeds. However, existing compression-oriented approaches, such as key-value (KV) cache compression and context compression, often either make an early decision about which contextual information to keep or rely on an external compressor. Such designs make it difficult to adapt the compressed context to the evidence needed by later reasoning steps. This paper introduces PReM (Preserve and Refresh Memory), a context-compression framework that maintains the long context as the model's internal layer-wise KV memory and learns what to preserve and when to refresh it. Specifically, PReM uses a dedicated memory layer to make memory-selection decisions, and a special memory token <m> to trigger refreshes during generation. To train this behavior, PReM introduces Phase-Separated Refresh Training, aligning memory selection with memory-conditioned generation while preserving continuity across refreshes. Experiments with 32K-token contexts show that PReM outperforms strong baselines under both 16x and 32x compression, while maintaining a favorable balance between answer quality and inference efficiency.
Repeated queries over shared documents repeatedly execute the same lower transformer layers. We introduce CoMem, which makes split depth j an explicit reusable-context axis: write one depth-j residual per token, select a bounded chunk set, and resume only layers [j:L). Among document-reuse systems we are aware of, CoMem jointly makes split depth a tunable serving axis and isolates it with a matched j=0 endpoint. On Qwen3-8B, j=12 reduces selected-pack Read from 931.9 to 664.4 ms (1.403x), with a 3.12-point RULER cost (95% CI [2.36, 3.93]); a continuous-prefix oracle recovers the full gap. The resulting depth axis quantifies a quality-latency-storage trade-off; a separate same-adapter, Write-inclusive pipeline is 2.74x faster. Equal-latency raw replay leads by 11.56 points with BM25, directly measuring an applicability boundary of prepaid depth rather than hiding it. CoMem stores 8 KiB/token versus 144 KiB/token for a protocol-aligned same-Qwen3 CacheBlend-style diagnostic; the cohorts and adaptation budgets are not matched. A context-position factorization identifies missing lower-layer document context as the dominant tested multikey error, and a 32-token overlap raises 92.5 to 98.5 without increasing persistent bytes or per-query Read. CoMem opens transformer depth as a measurable, tunable dimension for repeated-query long-context serving.