Default parameter. 6.5bpw and 8.0 bpw uses 8 bit lm_head layer, while 4.25bpw and 5.0bpw uses 6 bit lm_head layer.
(Layla (iOS / Android) support is in progress)
BeaverAI proudly presents...
Hubble 4B v1
Equipped with his five senses, man explores the universe around him and calls the adventure 'Science'.
image/pngDescription
This is a finetune of Nvidia's Llama 3.1 4B Minitron - a shrunk down model of Llama 3.1 8B 128K.
Usage
- ChatML or Text Completion
- Add
<|im_end|> as a stop token
Links
Technical Note
Hubble was trained on ChatML with <|end_of_text|> as the EOS token. If you encounter any issues with the model, please let me know!
