{
  "quant_method": "orbitquant",
  "weight_bits": 4,
  "activation_bits": 8,
  "rotation": "rpbh",
  "rotation_seed": 0,
  "block_size": "paper",
  "rotation_rounds": "auto",
  "codebook": "lloyd_max",
  "codebook_version": 3,
  "codebook_dtype": "float32",
  "row_norm_dtype": "bfloat16",
  "activation_norm_dtype": "float32",
  "activation_eps": 1e-10,
  "activation_quantizer": "absmax_int8",
  "weight_pack_dtype": "uint8",
  "target_policy": "universal",
  "target_policy_version": 2,
  "adaln_policy": "int4_rtn",
  "adaln_group_size": 64,
  "adaln_scale": "absmax",
  "output_gain_correction": false,
  "modules_to_convert": [
    "transformer_blocks.*.attn.to_q",
    "transformer_blocks.*.attn.to_k",
    "transformer_blocks.*.attn.to_v",
    "transformer_blocks.*.attn.to_out.0",
    "transformer_blocks.*.ff.net.0.proj",
    "transformer_blocks.*.ff.net.2",
    "token_refiner.refiner_blocks.*.attn.to_q",
    "token_refiner.refiner_blocks.*.attn.to_k",
    "token_refiner.refiner_blocks.*.attn.to_v",
    "token_refiner.refiner_blocks.*.attn.to_out.0",
    "token_refiner.refiner_blocks.*.ff.net.0.proj",
    "token_refiner.refiner_blocks.*.ff.net.2"
  ],
  "modules_to_use_adaln": [],
  "modules_to_not_convert": [],
  "modules_dtype_dict": {},
  "outlier_channels": {},
  "artifact_format_version": 1,
  "runtime_mode": "auto_fused",
  "activation_kernel_backend": "auto",
  "packed_matmul_block_m": 64,
  "packed_matmul_block_n": 64,
  "packed_matmul_block_k": 128,
  "packed_matmul_num_warps": 4,
  "weight_row_tile_size": 256,
  "w4a4_int8_weight_cache": false,
  "lowbit_boundary_protection": "auto",
  "lowbit_protected_blocks": 4,
  "lowbit_protected_bits": 4,
  "lowbit_interior_protection": "auto",
  "fused_layout": null
}