This repository contains the δ-mem TSW adapter for Qwen/Qwen3-4B-Instruct-2507, as presented in the paper δ-mem: Efficient Online Memory for Large Language Models. δ-mem is a lightweight online memory mechanism that augments a frozen backbone with a compact associative memory state. It projects context into a low-dimensional space and updates a state matrix via delta-rule learning, allowing for efficient long-term memory utilization without full fine-tuning or context extension. — Paper: https://huggingface.co/papers/2605.12357 — Repository: https://github.com/declare-lab/delta-Mem To use this adapter, you must first install the $delta$-mem codebase: This is not a standard PEFT LoRA adapter and should not be loaded with PeftModel or merged with mergeandunload(). δ-mem requires its specific runtime memory write/read path. To use this adapter, load the frozen base model, attach the modules, and then load the adapter weights using the provided codebase.
Модальности:
Генерация текста
Области применения:
Следование инструкциям
Задача: Генерация текста
Автор: declare-lab
Теги: delta-mem, memory-augmented-language-model, adapter, qwen3, long-context, test-time-memory, en
Лайков: 4 | Загрузок: 0
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
