#396 Add custom & auto quantization (#397)

README.md

@@ -188,6 +188,87 @@ from netspresso import NetsPresso
 netspresso = NetsPresso(email="YOUR_EMAIL", password="YOUR_PASSWORD")
 ```
 
+### ⭐⭐⭐ (New Feature) Quantizer ⭐⭐⭐
+
+#### Automatic quantization
+
+To start quantize a model, enter the model path, dataset path, and the desired quantization precision.
+
+The quantized model will be saved to the specified output directory (`output_dir`).
+
+```python
+from netspresso.enums import QuantizationPrecision, SimilarityMetric
+
+# 1. Declare quantizer
+quantizer = netspresso.quantizer()
+
+# 2. Run automatic quantization
+quantization_result = quantizer.automatic_quantization(
+    input_model_path="./examples/sample_models/test.onnx",
+    output_dir="./outputs/quantized/automatic_quantization",
+    dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
+    weight_precision=QuantizationPrecision.INT8,
+    activation_precision=QuantizationPrecision.INT8,
+    threshold=0,
+)
+```
+
+#### Custom precision quantization by layer name
+
+This method enables you to apply precision settings tailored to each layer, based on the recommendations, to optimize model.
+
+Or, you can modify it to your desired precision and optimize it.
+
+```python
+from netspresso.enums import QuantizationPrecision
+
+# 1. Declare quantizer
+quantizer = netspresso.quantizer()
+
+# 2. Recommendation precision
+metadata = quantizer.get_recommendation_precision(
+    input_model_path="./examples/sample_models/test.onnx",
+    output_dir="./outputs/quantized/recommendation",
+    dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
+    weight_precision=QuantizationPrecision.INT8,
+    activation_precision=QuantizationPrecision.INT8,
+    threshold=0,
+)
+recommendation_precisions = quantizer.load_recommendation_precision_result(metadata.recommendation_result_path)
+
+# 2. Run quantization by layer name
+quantization_result = quantizer.custom_precision_quantization_by_layer_name(
+    input_model_path="./examples/sample_models/test.onnx",
+    output_dir="./outputs/quantized/custom_precision_quantization_by_layer_name",
+    precision_by_layer_name=recommendation_precisions.layers,
+    dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
+)
+```
+
+#### Automatic quantization
+
+To start quantize a model, enter the model path, dataset path, and the desired quantization precision.
+
+The quantized model will be saved to the specified output directory (`output_dir`).
+
+```python
+from netspresso.enums import QuantizationPrecision, SimilarityMetric
+
+# 1. Declare quantizer
+quantizer = netspresso.quantizer()
+
+# 2. Run automatic quantization
+compression_result = quantizer.automatic_quantization(
+    input_model_path="./examples/sample_models/test.onnx",
+    output_dir="./outputs/quantized/automatic_quantization",
+    dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
+    weight_precision=QuantizationPrecision.INT8,
+    activation_precision=QuantizationPrecision.INT8,
+    threshold=0,
+)
+```
+
+
 ### Trainer
 
 #### Train

docs/conf.py

@@ -14,7 +14,7 @@
 project = "NetsPresso"
 copyright = "2023-2024, Nota Inc"
 author = "Nota AI"
-release = "1.7.0"
+release = "1.13.0"
 
 # -- General configuration ---------------------------------------------------
 # https://www.sphinx-doc.org/en/master/usage/configuration.html#general-configuration

docs/description.rst

@@ -6,6 +6,7 @@ API Description
 
    description/api/trainer/trainer
    description/api/compressor/compressor
+   description/api/quantizer/quantizer
    description/api/converter/converter
    description/api/benchmarker/benchmarker
    description/api/enums/enums

docs/description/api/enums/enum/onnx_operator.rst

@@ -0,0 +1,9 @@
+OnnxOperator
+############
+
+.. autoclass:: netspresso.enums.__init__.OnnxOperator
+    :members:
+    :undoc-members:
+    :member-order: bysource
+    :exclude-members: create_literal
+

docs/description/api/enums/enum/quantization_mode.rst

@@ -0,0 +1,9 @@
+QuantizationMode
+################
+
+.. autoclass:: netspresso.enums.__init__.QuantizationMode
+    :members:
+    :undoc-members:
+    :member-order: bysource
+    :exclude-members: create_literal
+

docs/description/api/enums/enum/quantization_precision.rst

@@ -0,0 +1,9 @@
+QuantizationPrecision
+#####################
+
+.. autoclass:: netspresso.enums.__init__.QuantizationPrecision
+    :members:
+    :undoc-members:
+    :member-order: bysource
+    :exclude-members: create_literal
+

docs/description/api/enums/enum/similarity_metric.rst

@@ -0,0 +1,9 @@
+SimilarityMetric
+################
+
+.. autoclass:: netspresso.enums.__init__.SimilarityMetric
+    :members:
+    :undoc-members:
+    :member-order: bysource
+    :exclude-members: create_literal
+

docs/description/api/enums/enums.rst

@@ -17,4 +17,8 @@ Enums
    enum/software_version
    enum/hardware_type
    enum/task_status
-   enum/data_type
+   enum/data_type
+   enum/quantization_mode
+   enum/quantization_precision
+   enum/similarity_metric
+   enum/onnx_operator

docs/description/api/quantizer/quantize/automatic_quantization.rst

@@ -0,0 +1,26 @@
+Automatic Quantization
+======================
+
+.. autofunction:: netspresso.quantizer.__init__.Quantizer.automatic_quantization
+
+
+Example
+-------
+
+.. code-block:: python
+
+    from netspresso import NetsPresso
+    from netspresso.enums import QuantizationPrecision
+
+
+    netspresso = NetsPresso(email="YOUR_EMAIL", password="YOUR_PASSWORD")
+
+    quantizer = netspresso.quantizer()
+    quantization_result = quantizer.automatic_quantization(
+        input_model_path="./examples/sample_models/test.onnx",
+        output_dir="./outputs/quantized/automatic_quantization",
+        dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
+        weight_precision=QuantizationPrecision.INT8,
+        activation_precision=QuantizationPrecision.INT8,
+        threshold=0,
+    )

docs/description/api/quantizer/quantize/custom_precision_quantization_by_layer_name.rst

@@ -0,0 +1,35 @@
+Custom Precision Quantization by Layer Name
+===========================================
+
+.. autofunction:: netspresso.quantizer.__init__.Quantizer.custom_precision_quantization_by_layer_name
+
+
+Example
+-------
+
+.. code-block:: python
+
+    from netspresso import NetsPresso
+    from netspresso.enums import QuantizationPrecision
+
+
+    netspresso = NetsPresso(email="YOUR_EMAIL", password="YOUR_PASSWORD")
+
+    quantizer = netspresso.quantizer()
+
+    recommendation_metadata = quantizer.get_recommendation_precision(
+        input_model_path="./examples/sample_models/test.onnx",
+        output_dir="./outputs/quantized/automatic_quantization",
+        dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
+        weight_precision=QuantizationPrecision.INT8,
+        activation_precision=QuantizationPrecision.INT8,
+        threshold=0,
+    )
+    recommendation_precisions = quantizer.load_recommendation_precision_result(recommendation_metadata.recommendation_result_path)
+
+    quantization_result = quantizer.custom_precision_quantization_by_layer_name(
+        input_model_path="./examples/sample_models/test.onnx",
+        output_dir="./outputs/quantized/custom_precision_quantization_by_layer_name",
+        dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
+        precision_by_layer_name=recommendation_precisions.layers,
+    )

docs/description/api/quantizer/quantize/custom_precision_quantization_by_operator_type.rst

@@ -0,0 +1,35 @@
+Custom Precision Quantization by Operator Type
+==============================================
+
+.. autofunction:: netspresso.quantizer.__init__.Quantizer.custom_precision_quantization_by_operator_type
+
+
+Example
+-------
+
+.. code-block:: python
+
+    from netspresso import NetsPresso
+    from netspresso.enums import QuantizationPrecision
+
+
+    netspresso = NetsPresso(email="YOUR_EMAIL", password="YOUR_PASSWORD")
+
+    quantizer = netspresso.quantizer()
+
+    recommendation_metadata = quantizer.get_recommendation_precision(
+        input_model_path="./examples/sample_models/test.onnx",
+        output_dir="./outputs/quantized/automatic_quantization",
+        dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
+        weight_precision=QuantizationPrecision.INT8,
+        activation_precision=QuantizationPrecision.INT8,
+        threshold=0,
+    )
+    recommendation_precisions = quantizer.load_recommendation_precision_result(recommendation_metadata.recommendation_result_path)
+
+    quantization_result = quantizer.custom_precision_quantization_by_operator_type(
+        input_model_path="./examples/sample_models/test.onnx",
+        output_dir="./outputs/quantized/custom_precision_quantization_by_operator_type",
+        dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
+        precision_by_operator_type=recommendation_precisions.operators,
+    )

docs/description/api/quantizer/quantize/recommendation_precision.rst

@@ -0,0 +1,27 @@
+Recommendation precision
+========================
+
+.. autofunction:: netspresso.quantizer.__init__.Quantizer.get_recommendation_precision
+
+
+Example
+-------
+
+.. code-block:: python
+
+    from netspresso import NetsPresso
+    from netspresso.enums import QuantizationPrecision
+
+
+    netspresso = NetsPresso(email="YOUR_EMAIL", password="YOUR_PASSWORD")
+
+    quantizer = netspresso.quantizer()
+    recommendation_metadata = quantizer.get_recommendation_precision(
+        input_model_path="./examples/sample_models/test.onnx",
+        output_dir="./outputs/quantized/automatic_quantization",
+        dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
+        weight_precision=QuantizationPrecision.INT8,
+        activation_precision=QuantizationPrecision.INT8,
+        threshold=0,
+    )
+    recommendation_precisions = quantizer.load_recommendation_precision_result(recommendation_metadata.recommendation_result_path)

docs/description/api/quantizer/quantize/uniform_precision_quantization.rst

@@ -0,0 +1,26 @@
+Plain Quantization
+======================
+
+.. autofunction:: netspresso.quantizer.__init__.Quantizer.uniform_precision_quantization
+
+
+Example
+-------
+
+.. code-block:: python
+
+    from netspresso import NetsPresso
+    from netspresso.enums import QuantizationPrecision, SimilarityMetric
+
+
+    netspresso = NetsPresso(email="YOUR_EMAIL", password="YOUR_PASSWORD")
+
+    quantizer = netspresso.quantizer()
+    quantization_result = quantizer.uniform_precision_quantization(
+        input_model_path="./examples/sample_models/test.onnx",
+        output_dir="./outputs/quantized/uniform_precision_quantization",
+        dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
+        metric=SimilarityMetric.SNR,
+        weight_precision=QuantizationPrecision.INT8,
+        activation_precision=QuantizationPrecision.INT8,
+    )

docs/description/api/quantizer/quantizer.rst

@@ -0,0 +1,15 @@
+Quantizer
+==========
+
+
+Quantize
+--------
+
+.. toctree::
+   :maxdepth: 1
+
+   quantize/uniform_precision_quantization
+   quantize/automatic_quantization
+   quantize/recommendation_precision
+   quantize/custom_precision_quantization_by_layer_name
+   quantize/custom_precision_quantization_by_operator_type

examples/quantizer/automatic_quantization.py

@@ -0,0 +1,29 @@
+from pathlib import Path
+
+from netspresso import NetsPresso
+from netspresso.enums import QuantizationPrecision, SimilarityMetric
+
+
+EMAIL = "YOUR_EMAIL"
+PASSWORD = "YOUR_PASSWORD"
+
+netspresso = NetsPresso(email=EMAIL, password=PASSWORD)
+
+# 1. Declare quantizer
+quantizer = netspresso.quantizer()
+
+# 2. Set variables for quantize
+input_model = "./examples/sample_models/test.onnx"
+OUTPUT_DIR = "./outputs/quantized/automatic_quantization"
+CALIBRATION_DATASET_PATH = "./examples/sample_datasets/pickle_calibration_dataset_128x128.npy"
+
+# 3. Automatic Quantization
+quantization_result = quantizer.automatic_quantization(
+    input_model_path=input_model,
+    output_dir=f"{OUTPUT_DIR}/{Path(input_model).stem}",
+    dataset_path=CALIBRATION_DATASET_PATH,
+    weight_precision=QuantizationPrecision.INT8,
+    activation_precision=QuantizationPrecision.INT8,
+    metric=SimilarityMetric.SNR,
+    threshold=0,
+)

examples/quantizer/custom_precision_quantization_by_layer_name.py

@@ -0,0 +1,38 @@
+from pathlib import Path
+
+from netspresso import NetsPresso
+from netspresso.enums import QuantizationPrecision
+
+
+EMAIL = "YOUR_EMAIL"
+PASSWORD = "YOUR_PASSWORD"
+
+netspresso = NetsPresso(email=EMAIL, password=PASSWORD)
+
+
+# 1. Declare quantizer
+quantizer = netspresso.quantizer()
+
+# 2. Set variables for quantize
+input_model = "./examples/sample_models/test.onnx"
+OUTPUT_DIR = "./outputs/quantized/onnx2onnx_2"
+CALIBRATION_DATASET_PATH = "./examples/sample_datasets/pickle_calibration_dataset_128x128.npy"
+
+# 3-1. Recommendation precision
+recommendation_metadata = quantizer.get_recommendation_precision(
+    input_model_path=input_model,
+    output_dir=f"{OUTPUT_DIR}/{Path(input_model).stem}",
+    dataset_path=CALIBRATION_DATASET_PATH,
+    weight_precision=QuantizationPrecision.INT8,
+    activation_precision=QuantizationPrecision.INT8,
+    threshold=0,
+)
+recommendation_precisions = quantizer.load_recommendation_precision_result(recommendation_metadata.recommendation_result_path)
+
+# 3-2. Custom Quantization
+quantization_result = quantizer.custom_precision_quantization_by_layer_name(
+    input_model_path=input_model,
+    output_dir=f"{OUTPUT_DIR}/{Path(input_model).stem}",
+    precision_by_layer_name=recommendation_precisions.layers,
+    dataset_path=CALIBRATION_DATASET_PATH,
+)