config.json

{
  "model_type": "qwen3-asr-text-decoder-anemll",
  "source_model": "Qwen/Qwen3-ASR-0.6B",
  "converter": "ANEMLL 0.3.5 (github.com/Anemll/Anemll f4ad26d)",
  "context_length": 512,
  "batch_size": 64,
  "hidden_size": 1024,
  "num_layers": 28,
  "vocab_size": 151936,
  "lm_head_slices": 16,
  "quantization": "LUT8, 8 channels per group (decoder, lm_head)",
  "files": {
    "decoder": "decoder.mlmodelc (functions: infer = 1 position, prefill = 64 positions; one shared KV state)",
    "lm_head": "lm_head.mlmodelc (argmax per vocabulary slice)"
  }
}