experiment_name: production
output_dir: ./outputs/production

model:
  d_model: 512
  n_heads: 8
  n_layers: 6
  ff_multiplier: 4
  max_seq_len: 256
  vocab_size: 30000
  dropout: 0.1
  tt_rank: 16
  tt_min_rank: 4
  use_tensor_ffn: true
  n_qubits: 6
  n_quantum_layers: 3
  quantum_sparsity: 0.8
  use_quantum: true
  rank_alpha: 2.0
  rank_smoothing: 0.95

training:
  learning_rate: 2.0e-4
  weight_decay: 0.01
  warmup_steps: 500
  max_epochs: 15
  batch_size: 4
  gradient_accumulation_steps: 4
  max_grad_norm: 1.0
  seed: 42
  lr_scheduler: cosine
  lr_min_factor: 0.05

budget:
  max_params: 50000000
  max_latency_ms: 50.0
  max_energy_per_query: 500.0
  target_compression_ratio: 2.0