transformer/quantization_config.json

{
  "add_skip_keys": false,
  "dequantize_fp32": false,
  "dynamic_loss_threshold": 0.01,
  "group_size": 0,
  "is_integer": true,
  "is_training": false,
  "modules_dtype_dict": {},
  "modules_to_not_convert": [
    "prediction_coefs",
    "siglip_embedder",
    "cap_embedder",
    "norm_out",
    "correction_coefs",
    "lm_head",
    "embed_tokens",
    "layers.0.adaLN_modulation.0.weight",
    "all_final_layer",
    "embedding_projection",
    "t_embedder",
    "proj_out",
    "all_x_embedder"
  ],
  "non_blocking": false,
  "quant_conv": false,
  "quant_method": "sdnq",
  "quantization_device": null,
  "quantized_matmul_dtype": null,
  "return_device": null,
  "sdnq_version": "0.1.3",
  "svd_rank": 32,
  "svd_steps": 8,
  "use_dynamic_quantization": false,
  "use_grad_ckpt": true,
  "use_quantized_matmul": true,
  "use_quantized_matmul_conv": false,
  "use_static_quantization": true,
  "use_stochastic_rounding": false,
  "use_svd": false,
  "weights_dtype": "int8"
}