# model/tokenizer
model_name: "zpn/llama-7b"
tokenizer_name: "zpn/llama-7b"
gradient_checkpointing: true

# dataset
streaming: false
num_proc: 64
dataset_path: "data.jsonl"
max_length: 512
batch_size: 32

# train dynamics
lr: 5.0e-5
eval_every: 2000
eval_steps: 100
save_every: 2000
output_dir: "ckpts/llama-7b"
checkpoint: null
lora: false
warmup_steps: 100

# logging
wandb: false
wandb_entity: zanussbaum
wandb_project: llama
seed: 42