test-time-training/ttt-lm-pytorch
PyTorch implementation of Test-Time Training (TTT) layers, a new class of sequence modeling layers that use self-supervised learning to update their hidden states at test time.
Not currently ranked — collecting fresh signals.
star history
The repository implements TTT-Linear and TTT-MLP layers that replace traditional RNN hidden states with machine learning models themselves. The hidden state is updated through self-supervised learning steps during both training and inference, achieving linear complexity with expressive hidden states for long-context sequence modeling. The implementation is based on Huggingface Transformers for easy model loading and text generation.
Frequently asked
- What is test-time-training/ttt-lm-pytorch?
- PyTorch implementation of Test-Time Training (TTT) layers, a new class of sequence modeling layers that use self-supervised learning to update their hidden states at test time.
- Is ttt-lm-pytorch open source?
- Yes — test-time-training/ttt-lm-pytorch is open source, released under the MIT license.
- What language is ttt-lm-pytorch written in?
- test-time-training/ttt-lm-pytorch is primarily written in Python.
- How popular is ttt-lm-pytorch?
- test-time-training/ttt-lm-pytorch has 1.4k stars on GitHub.
- Where can I find ttt-lm-pytorch?
- test-time-training/ttt-lm-pytorch is on GitHub at https://github.com/test-time-training/ttt-lm-pytorch.