# hydra:
#   run:
#     dir: ./my_output_dir

root_dir: ${hydra:runtime.cwd}

eval:
  save_path: checkpoints/cs336_lm_owt/
  iteration: 5000
  prompt: That empowerment comes with added scrutiny
  max_new_tokens: 64
  temperature: 1.0
  top_k: 50

model_type: cs336_lm
model:
  vocab_size: 32000
  context_length: 256
  d_model: 512
  num_layers: 4
  num_heads: 16
  d_ff: 1344
  rope_theta: 10000.0

dataset_name: owt     # `TinyStories` or `owt`, change as needed
dataset_split: valid          # `train` or `valid`, change as needed
tokenizer_dir: ${root_dir}/tokenizers/${dataset_name}_${dataset_split}/ # save path for tokenizer
tokenizer:
  vocab_path: ${tokenizer_dir}/vocab.pkl
  merges_path: ${tokenizer_dir}/merges.pkl
  special_tokens: ["<|endoftext|>"]