hlwq_config.json

{
  "quantization_method": "PolarQuant",
  "version": "v5_vision",
  "weight_format": "torchao_int4",
  "weight_bits": 4,
  "vision_weight_bits": 16,
  "kv_cache_bits": 3,
  "block_size": 128,
  "head_dim": 256,
  "base_model": "TeichAI/gemma-4-31B-it-Claude-Opus-Distill",
  "multimodal": true,
  "pipeline_tag": "image-text-to-text",
  "num_layers": 60,
  "num_kv_heads": 16,
  "vram_gb": 22.8,
  "download_gb": 21.8,
  "tok_s": 24.9
}