If you want to use huggingface's llama or other LLMs, you must have permission to import models into your huggingface account, and you can do so by logging in with the code huggingface-cli login
in your command before using them.
Also, because the code uses wandb to track your learning progress, you will need to enter your wandb account and token information via wandb login
to use the
transformers == 4.39.3
pytorch == 2.3.0+cu118
peft == 0.10.0
trl == 0.8.6
bitsandbytes == 0.43.1
wandb == 0.16.6
python run.py [-h] [-m MODEL_NAME] [-c CACHE_DIR] [-lr LEARNING_RATE] [-e NUM_EPOCHS] [-ml MAX_LEN]
[-trn TRAIN_PATH] [-val VAL_PATH] [-tst TEST_PATH] [-sp SAVE_PATH] [-r R]
[-ld LORA_DROPOUT] [-la LORA_ALPHA] [--seed SEED]
python run.py -m save_model/Llama-2-13b-chat-hf -lr 3e-4 -e 300 -r 16 -trn data/english-train.json -val data/english-dev.json -tst data/english-test.json -sp checkpoints