{
  "version": "v2",
  "decoder_type": "semi_autoregressive",
  "vocab_size": 50304,
  "dim": 768,
  "concept_dim": 384,
  "num_concepts": 1024,
  "num_quantizers": 4,
  "downsample_rate": 4,
  "planner_layers": 5,
  "decoder_layers": 5,
  "num_heads": 8,
  "dropout": 0.0,
  "max_seq_len": 1024,
  "rope_theta": 10000.0
}