hlwq_config.json

{
  "quantization_method": "PolarQuant",
  "version": "v5",
  "weight_bits": 4,
  "kv_cache_bits": 3,
  "block_size": 128,
  "head_dim": 128,
  "base_model": "Jackrong/Qwen3.5-9B-Neo",
  "num_layers": 32,
  "num_kv_heads": 4,
  "vram_gb": 6.3,
  "tok_s": 42.8
}