This is quantized version of driaforall/mem-agent created using llama.cpp Based on Qwen3-4B-Thinking-2507, this model was trained using GSPO (Zheng et al., 2025) over an agent scaffold that is built around an Obisidian-like memory system and the tools required to interact with it. The model was trained on the following subtasks: — Retrieval: Retrieving relevant information when needed from the memory system. In this subtask, we also trained the model on filtering the retrieved information and/or obfuscating it completely. — Updating: Updating the memory system with new information. — Clarification: Asking for clarification when the user query is not clear/contradicting with the information in the memory system. In the scaffold, the model uses , and tags to structure its response. Using only when it’s done interacting with the memory. The block is executed in a sandbox with the tools and the results of the code block are returned in a tag to the model, forming the agentic loop. The model is also trained to be able to handle optional filters given by the user in between tags after the user query. These filters are used to filter the retrieved information and/or obfuscate it…
Модальности:
Генерация текста
Задача: Генерация текста
Автор: QuantFactory
Теги: gguf, endpoints_compatible
Лайков: 3 | Загрузок: 1,156
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
