model: mlx-community/Qwen2.5-7B-Instruct-4bit
train: true
data: data/train
adapter_path: adapters/buddhagpt-v1
batch_size: 1
grad_accumulation_steps: 8
iters: 1200
learning_rate: 1e-5
num_layers: 16
lora_parameters:
  rank: 16
  scale: 20.0
  dropout: 0.05
max_seq_length: 1024
steps_per_eval: 200
save_every: 200