The same goal as neurons.c but in python; cuz "damn C got hands"
Use Python as a proxy to achieve the goals of neurons.c (https://github.com/DaertML/neurons.c). The struggle to port Mistral with the code from llama2.c (or any of the alternatives) is starting to wear me off. I learned a ton of C in the last days (mainly because I am not proficient on it); but we need to accelerate and get practical.
- Mistral 7B: ✔️
This code is possible thanks to the repo https://github.com/naklecha/llama3-from-scratch Most of the code is copy paste, changed the names of the layers, and needed to use AutoTokenizer to make a tiny mistral to work.