config.json

{
  "model": "SPRINGLab/Indic-Mio",
  "codec": "Aratako/MioCodec-25Hz-24kHz",
  "format": "litert",
  "variant": "fp16",
  "loader": "stock transformers Qwen3ForCausalLM",
  "sample_rate": 24000,
  "samples_per_token": 960,
  "vocab_size": 164480,
  "speech_token_offset": 151669,
  "speech_token_count": 12800,
  "eos_token_ids": [
    151645,
    151643
  ],
  "pad_token_id": 151643,
  "prompt_template": "<|im_start|>user\n{text}<|im_end|>\n<|im_start|>assistant\n",
  "emotion_markers": [
    "<happy>",
    "<sad>",
    "<angry>",
    "<disgust>",
    "<fear>",
    "<surprise>"
  ],
  "max_cache_len": 512,
  "prefix_len": 64,
  "decoder_tokens": 384,
  "ref_samples": 240000,
  "decoder_head": "spec",
  "decoder_dynamic": false,
  "decoder_masked": true,
  "decoder_max_tokens": 384,
  "graphs": {
    "text_prefill": {
      "file": "indicmio-text-prefill.tflite",
      "precision": "fp16"
    },
    "token_step": {
      "file": "indicmio-token-step.tflite",
      "precision": "fp16"
    },
    "audio_decoder": {
      "file": "indicmio-audio-decoder.tflite",
      "precision": "fp32"
    },
    "ref_encoder": {
      "file": "indicmio-ref-encoder.tflite",
      "precision": "fp32"
    }
  }
}