MemTensor/MemT-1.0-Nano
AI model · https://huggingface.co/MemTensor/MemT-1.0-Nano
First seen by Oriel on .
AI summary
MemT-1.0-Nano is a memory-augmented language model that retains information in persistent latent state, reducing storage needs while supporting context-free memory retrieval.
- Its Native Memory Block combines evolving local memory with trainable hyper memory, updating state at the end of each turn or segment.
- At 128K history length, persistent state uses approximately 66× less storage than a full-context KV cache.
- MemT-1.0-Nano has 36B total parameters, 3B activated parameters, 451M trainable memory parameters, and a 10M dynamic memory state.
- Across 11 memory benchmarks, it ranks second among evaluated parametric-memory methods; textual-memory systems perform better overall, especially for detailed retrieval from long histories.
- The model supports saving and resuming latent memory without restoring plaintext conversation, and its weights are released under Apache License, Version 2.0.
#large language models #model memory #agentic coding
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Likes
Total likes reported by the source.
- Hugging Face Models: 143, latest observation ; +25 over observed period since
Downloads (30d)
Rolling 30-day downloads. A change in this value is not new downloads during the observed period.
- Hugging Face Models: 30,338, latest observation ; +11,573 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Models | MemTensor/MemT-1.0-Nano | https://huggingface.co/MemTensor/MemT-1.0-Nano Read discussion |