basically AI
Pebble is our line of small, efficient language models: base models for further training, and instruction-tuned chat variants built on top of them. All Pebble models are released openly on Hugging Face.
| Model | Parameters | Status | Context | Type | Non-CUDA compatible | Link |
|---|---|---|---|---|---|---|
| Pebble 25M | 25M | Recommended | 512 | Base | No | Hugging Face ↗ |
| Pebble 10M | 10M | Recommended | 512 | Base | No | Hugging Face ↗ |
| Pebble 25M Chat | 25M | Current | 512 | Basic IFT | No | Hugging Face ↗ |
| Pebble 10M Chat | 10M | Current | 512 | Basic IFT | No | Hugging Face ↗ |
These models require a CUDA-capable NVIDIA GPU. They do not currently run on CPU, AMD, or Intel hardware.