{ "fp16": {}, "q8": { "per_model_config": { "model": { "op_types": [ "Abs", "Add", "And", "ArgMax", "BatchNormalization", "Cast", "Clip", "Concat", "Conv", "Div", "Equal", "Exp", "Expand", "Gather", "GatherND", "Greater", "Less", "Max", "MaxPool", "Min", "Mul", "Not", "Pow", "Range", "Reciprocal", "ReduceMax", "ReduceSum", "Relu", "Reshape", "Resize", "Shape", "Sigmoid", "Size", "Slice", "Split", "Squeeze", "Sub", "Tile", "TopK", "Transpose", "Unsqueeze" ], "weight_type": "QUInt8" } }, "per_channel": false, "reduce_range": false }, "int8": { "per_model_config": { "model": { "op_types": [ "Abs", "Add", "And", "ArgMax", "BatchNormalization", "Cast", "Clip", "Concat", "Conv", "Div", "Equal", "Exp", "Expand", "Gather", "GatherND", "Greater", "Less", "Max", "MaxPool", "Min", "Mul", "Not", "Pow", "Range", "Reciprocal", "ReduceMax", "ReduceSum", "Relu", "Reshape", "Resize", "Shape", "Sigmoid", "Size", "Slice", "Split", "Squeeze", "Sub", "Tile", "TopK", "Transpose", "Unsqueeze" ], "weight_type": "QInt8" } }, "per_channel": false, "reduce_range": false }, "uint8": { "per_model_config": { "model": { "op_types": [ "Abs", "Add", "And", "ArgMax", "BatchNormalization", "Cast", "Clip", "Concat", "Conv", "Div", "Equal", "Exp", "Expand", "Gather", "GatherND", "Greater", "Less", "Max", "MaxPool", "Min", "Mul", "Not", "Pow", "Range", "Reciprocal", "ReduceMax", "ReduceSum", "Relu", "Reshape", "Resize", "Shape", "Sigmoid", "Size", "Slice", "Split", "Squeeze", "Sub", "Tile", "TopK", "Transpose", "Unsqueeze" ], "weight_type": "QUInt8" } }, "per_channel": false, "reduce_range": false }, "q4": { "block_size": 32, "is_symmetric": true, "accuracy_level": null }, "bnb4": { "block_size": 64, "quant_type": 1 } }