PowerUp-LLM

An LLM optimizer that streamlines matrix vector multiplication process by implementing multithreading and synchronization techniques. This optimizer is compatible for Linux machines.

Improvements

(see folder "examples")

1. Running the matrix vector multiplication using a single thread. Time taken: 4.927 s

2. Running the matrix vector multiplication using 4 threads. Time taken: 2.338 s

3. Running the matrix vector multiplication using 16 threads. Time taken: 1.548 s

Running the script

1. Clone the repository

git clone https://github.com/davinnev/PowerUp-LLM.git

2. Navigate to the source folder

cd src

3. Compile the code

gcc llama2.c -o llama2

4. Try running the optimizer

./llama2 {seed} {num of threads}

For example,

./llama2 42 4