MemT-1.0 is a new generation of memory foundation models that makes persistent, dynamically evolving memory an intrinsic part of model computation.
Why it matters now: MemT-1.0 is a new generation of memory foundation models that makes persistent, dynamically evolving memory an intrinsic part of model computation.
MemT-1.0 integrates dynamic local memory and trainable hyper memory to retain and leverage historical information without accessing original context at inference. The project makes three main contributions: memory architecture, training and evaluation, and scaling and applications. MemT-1.0 introduces an agentic framework for reusing procedural experience in long-horizon coding tasks. The model has been evaluated on 11 memory benchmarks, ranking first and second among the evaluated parametric-memory methods. MemT-1.0 has two main models: MemT-1.0-Nano and MemT-1.0-Flash, with different backbone architectures and parameter counts. The model can be used for a variety of tasks, including conversational AI and coding tasks. The project provides a Python client for interacting with the model, as well as a set of pre-trained models and evaluation scripts.
On a Linux server with NVIDIA CUDA and Python 3.11 or newer, install from the repository's `inference` directory