speech_token_meta.json

{
  "model_name_or_path": "./outputs/llama_st_to_t_clean100_freeze_emb",
  "text_vocab_size": 128256,
  "s_id": 128256,
  "t_id": 128257,
  "hubert_offset": 128258,
  "total_vocab_size": 128758,
  "codebook_size": 500,
  "patch_mode": "align",
  "patch_size": 4,
  "max_patch_length": 32,
  "pooling": "mean",
  "local_layers": 1,
  "local_heads": 8,
  "cross_heads": 8,
  "bidirectional_local_encoder": false,
  "input_embeddings_frozen": true,
  "freeze_lm_head": false,
  "objective": "speech_patches_to_text_only",
  "patch_encoder_state": "speech_patch_encoder.pt"
}