Instructions to use navthings/airgpt-night with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use navthings/airgpt-night with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download navthings/airgpt-night --local-dir airgpt-night
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
airgpt-night
a 40M param model trained from scratch in eight hours on a macbook air (m5, 16gb, no fan) with airgpt. the air didn't throttle once the whole night.
it writes longer stories that mostly hold together. the chat is still pretty weak, it loops and makes things up.
- pretrain val loss 2.11, chat val loss 1.62
- 120M tokens of tinystories + cosmopedia (50/50), then short chats from everyday-conversations and smol-smoltalk
model.safetensorsis the chat model,base.safetensorsis the model before chat training
to talk to it:
git clone https://github.com/navthings/airgpt && cd airgpt
pip install -r requirements.txt
hf download navthings/airgpt-night --local-dir runs/night
python chat.py runs/night
python chat.py runs/night --base uses the base model, which just continues whatever you type.
- Downloads last month
- 17
Model size
39.6M params
Tensor type
F32
ยท
Hardware compatibility
Log In to add your hardware
Quantized
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support