ort_config.json

{
  "opset": 17,
  "optimization": {
    "enable_gelu_approximation": false,
    "enable_gemm_fast_gelu": false,
    "enable_layer_norm": true,
    "enable_skip_layer_norm": true,
    "enable_bias_skip_layer_norm": true,
    "enable_bias_gelu": true,
    "enable_gelu": true,
    "enable_attention": true,
    "use_multi_head_attention": true,
    "enable_skip_simplified_layer_norm": true
  },
  "quantization": {
    "format": "QOperator",
    "mode": "IntegerOps",
    "per_channel": false,
    "reduce_range": true,
    "weight_type": "QInt8"
  }
}