example/7B.yaml

# data
data:
  instruct_data: "/root/data/mol_instructions_train.jsonl"  # Fill this with the path to your training data
  data: ""  # Optionally fill with pretraining data
  eval_instruct_data: ""  # Optionally fill with evaluation data

# model
model_id_or_path: "/root/mistral_models/7B-v0.3"  # Path to downloaded model
lora:
  rank: 64

# optim
seq_len: 32768
batch_size: 2
 #TODO try other values
max_steps: 500
optim:
  lr: 5.e-5
  weight_decay: 0.05
  pct_start: 0.05

# other
seed: 99
log_freq: 1
eval_freq: 100
no_eval: True
ckpt_freq: 100

ckpt_only_lora: False  # Save only trained LoRA adapters. Set to `False` to merge LoRA adapter into the base model and save full fine-tuned model

run_dir: "/root/mistral-finetune/runseed99"

wandb:
  project: "CHEMISTral7b-ft"
  offline: False  # Set to True if you want to use wandb in offline mode
  key: "aaf77f83a4e316f6a8b47fa975ab6b5e73c7c8df"  # Optionally set your WandB API key
  run_name: "runseed99"  # Optionally name your WandB run