config.json

{
  "architectures": [
    "GlapModel"
  ],
  "auto_map": {
    "AutoConfig": "configuration_glap.GlapConfig",
    "AutoModel": "modeling_glap.GlapModel"
  },
  "model_type": "glap",
  "audio_embed_dim": 768,
  "audio_depth": 12,
  "audio_num_heads": 12,
  "patch_size": [
    64,
    4
  ],
  "patch_stride": [
    64,
    4
  ],
  "target_length": 1008,
  "sample_rate": 16000,
  "text_vocab_size": 256206,
  "text_model_dim": 1024,
  "text_num_layers": 24,
  "text_num_heads": 16,
  "text_ffn_inner_dim": 8192,
  "text_max_seq_len": 514,
  "text_pad_idx": 0,
  "text_dropout_p": 0.1,
  "embed_size": 1024
}