Tag
Titan-Llama 提出一种后训练方法，通过分层记忆架构和测试时参数更新，将预训练注意力模型改造为具备外部神经记忆模块的长上下文高效推理模型，在 QuALITY 和 LongHealth 上显著提升准确率并降低推理延迟与显存占用。
Remory 提出一种神经记忆网络，用有限的软记忆 token 补充文本摘要，使冻结的 LLM 在仅用 5.2% 输入位置的情况下逼近全上下文表现，并在 BrowseComp 和 Terminal-Bench 2.1 等长时程 agent 基准上带来一致提升、显著减少重复工具输出和错误。
REMORY introduces a neural memory network that supplements textual summaries with a bounded sequence of soft memory tokens for context compaction in long-horizon agents, enabling a frozen LLM to approach full-context performance using only 5.2% of input positions with consistent gains on Qwen3.8-27B and GLM-5.3-Flash.
MoNe is a lightweight modular neural memory that attaches to frozen pretrained Transformers to enable efficient long-context inference without retraining, reducing compute and memory usage significantly.