gpt4all/gpt4all-training/configs/train/finetune_openllama.yaml

# model/tokenizer
model_name: "openlm-research/open_llama_7b"
tokenizer_name: "openlm-research/open_llama_7b"
gradient_checkpointing: true
save_name: "nomic-ai/gpt4all-openllama"

# dataset
streaming: false
num_proc: 64
dataset_path: "nomic-ai/gpt4all-updated"
revision: null
max_length: 1024
batch_size: 32

# train dynamics
lr: 2.0e-5
min_lr: 0 
weight_decay: 0.0
eval_every: 500
log_every: 10
save_every: 1000
log_grads_every: 500
output_dir: "ckpts/falcon"
checkpoint: null
lora: false
warmup_steps: 500
num_epochs: 3 

# logging
wandb: true
wandb_entity: "gpt4all"
wandb_project_name: "gpt4all"
seed: 42