Adaptation of Llama2.c to run on a PULP microcontroller.
Clone this project
git clone https://github.com/argnaan/pulp_llama2.gitGet PULP-TrainLib from its github repository
git clone https://github.com/pulp-platform/pulp-trainlib.gitAnd modify Makefile with the correct path to the Train-Lib
Generate the header file which contains the configuration and the weights of the model:
make get_header
It can be runned thorugh GVSoC, a PULP chip simulator, part of the pulp-sdk.
make clean all run
Look inside the genHeader.c file to modify some option such as the initial prompt or the number of steps, then regenerate the header. You can also try different model in the format of the Llama2.c project.