config.json

{
  "architectures": [
    "SpeechTextMultiTask"
  ],
  "bos_token_id": 1,
  "encoder_config": {
    "attn_dropout": 0.1,
    "attn_flash": true,
    "conv_dropout": 0.1,
    "conv_expansion_factor": 2,
    "conv_kernel_size": 7,
    "depth": 6,
    "dim_head": 48,
    "ff_dropout": 0.1,
    "ff_mult": 4,
    "heads": 8
  },
  "eos_token_id": 2,
  "hidden_size": 384,
  "model_type": "speech_text_multitask",
  "num_speech_tokens": 1024,
  "num_text_tokens": 256,
  "pad_token_id": 0,
  "speech_tokenizer_name": null,
  "tokenizer_name": "byte",
  "torch_dtype": "bfloat16",
  "transformers_version": "4.44.2",
  "use_ctc_correction": true,
  "use_speech_mlm": true
}