File size: 2,017 Bytes
747451d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 | general:
logs_dir: logs
saved_models_dir: saved_models
display_figures: False
gpu_memory_limit: 24
model:
model_path: ../../stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/imagenet/mobilenetv1_a050_224/mobilenetv1_a050_224.keras
operation_mode: quantization
dataset:
dataset_name: imagenet
quantization_path: /local/data/ic_imagenet_2012/quantization/
quantization_split: 0.01
preprocessing:
rescaling:
scale: 1/127.5
offset: -1
resizing:
interpolation: nearest
aspect_ratio: fit
color_mode: rgb
quantization:
operating_mode: inspection
quantizer: Onnx_quantizer
quantization_type: PTQ
target_opset: 21
export_dir: quantized_models
quantization_input_type: uint8
quantization_output_type: float
onnx_quant_parameters:
weightType: Int4 # Int4, UInt4, Int8, UInt8, Int16, UInt16
activType: Int8 # Int4, UInt4, Int8, UInt8, Int16, UInt16
calibrate_method: MinMax # MinMax or Entropy
op_types_to_quantize: #['Conv'] for example or keep empty
nodes_to_quantize: #['Conv__229'] onnx layer names or keep empty, not compatible with weights_tensor_override
nodes_to_exclude: ['Conv__237'] # onnx layer names or keep empty
onnx_extra_options:
WeightSymmetric: true
ActivationSymmetric: false
QuantizeBias: true
CalibMovingAverage: true
weights_tensor_override: [['mobilenet_0.50_224_1/conv1_1/convolution/merged_input:0', {'quant_type': Int8, "axis": 0}],
['mobilenet_0.50_224_1/conv_pw_7_1/convolution/merged_input:0', {'quant_type': Int8, "axis": 0}]
]
activations_tensor_override: [['mobilenet_0.50_224_1/conv1_relu_1/Relu6:0', {'quant_type': Int16, 'scale': 0.0001, 'zero_point': 0}],
]
iterative_quant_parameters:
inspection_split: 0.8
mlflow:
uri: ./tf/src/experiments_outputs/mlruns
hydra:
run:
dir: ./tf/src/experiments_outputs/${now:%Y_%m_%d_%H_%M_%S} |