{
  "shard_idx": 0,
  "sequence_count": 128,
  "token_count": 15431,
  "hidden_states": 25,
  "hidden_size": 2048,
  "model": "Qwen/Qwen3.5-2B",
  "max_input_tokens": 128,
  "max_new_tokens": 16,
  "quantization": {
    "dtype": "int8",
    "scheme": "symmetric per-token per-hidden-state",
    "scale_dtype": "float16",
    "formula": "float ~= int8 * scale"
  }
}