runtime_manifest.json

{
  "model_family": "audio8_tts",
  "activation_dtype": "float32",
  "slow_logits_layout": "semantic_then_eos",
  "slow_logits_size": 4097,
  "kv_attention_layout": "valid_prefix",
  "max_seq_len": 2048,
  "num_layers": 24,
  "num_fast_layers": 4,
  "num_codebooks": 10,
  "n_local_heads": 2,
  "fast_n_local_heads": 2,
  "head_dim": 64,
  "fast_head_dim": 64,
  "fast_dim": 896,
  "vocab_size": 155776,
  "codebook_size": 4096,
  "semantic_begin_id": 151678,
  "semantic_end_id": 155773,
  "eos_token_id": 151645,
  "pad_token_id": 151643,
  "sample_rate": 44100,
  "codec_sample_rate": 44100,
  "codec_frame_size": 2048,
  "codec_hop_length": 2048,
  "ras_window_size": 10,
  "ras_top_p": 0.9,
  "ras_temperature": 1.0,
  "default_precision": "fp16",
  "available_precisions": [
    "fp16"
  ],
  "slow_models": {
    "fp16": "slow_ar_fp16.onnx"
  },
  "fast_models": {
    "fp16": "fast_ar_fp16.onnx"
  },
  "default_codec_precision": "fp32",
  "available_codec_precisions": [
    "fp32"
  ],
  "codec_models": {
    "fp32": "codec_decoder_fp32.onnx"
  }
}