manifest.json

{
  "name": "GigaAM v3 MLX",
  "bundle_id": "gigaam-v3-mlx",
  "source_model": "ai-sage/GigaAM-v3",
  "source_revision": "e2e_rnnt",
  "architecture": "conformer-rnnt",
  "model_type": "rnnt",
  "precision": "fp16",
  "runtime": "mlx",
  "runtime_target": "native-apple",
  "sample_rate": 16000,
  "channels": 1,
  "total_params": 222519937,
  "weights_file": "weights.fp16.safetensors",
  "tokenizer_file": "tokenizer.model",
  "tokenizer_vocab_file": "tokenizer_vocab.json",
  "blank_id": 1024,
  "vocab_size": 1024,
  "num_classes": 1025,
  "decoding": "greedy-rnnt",
  "frontend_files": {
    "hann_window": "hann_window.f32.bin",
    "mel_filterbank": "mel_filterbank_mel_freq.f32.bin"
  },
  "feature_config": {
    "sample_rate": 16000,
    "n_mels": 64,
    "n_fft": 320,
    "win_length": 320,
    "hop_length": 160,
    "center": false,
    "mel_scale": "htk",
    "mel_norm": null,
    "power": 2.0
  },
  "encoder_config": {
    "n_layers": 16,
    "d_model": 768,
    "n_heads": 16,
    "self_attention_model": "rotary",
    "pos_emb_max_len": 5000,
    "conv_kernel_size": 5,
    "subsampling": "conv1d",
    "subsampling_factor": 4,
    "conv_norm_type": "layer_norm"
  },
  "target_platforms": [
    "iOS",
    "iPadOS",
    "macOS Apple Silicon"
  ]
}