A small pet project fine-tuning Microsoft's Phi3.5-mini-instruct to summarize dialogue. Developed as part of a set of a benchmarking workflows I wrote for the C3 compute hardware upgrade.
Use the provided environment.yml file to set up the running environment using Conda.
conda env create -f environment.yml
conda activate phi3_dialogueThis script can run on multiple GPUs using the 🤗 Accelerate library. To launch distributed training, make sure to set up 🤗 Accelerate with accelerate config before calling main.py the first time.
After configuring multi-GPU/node acceleration launch the script from the command line with:
accelerate launch --num_processes <number of gpus> main.py {--args}