#120 Update example code for the changed interface (#124)

examples/compressor/automatic_compression.py

@@ -1,20 +1,24 @@
-from netspresso.compressor import Compressor
+from netspresso import NetsPresso
 
 
 EMAIL = "YOUR_EMAIL"
 PASSWORD = "YOUR_PASSWORD"
-compressor = Compressor(email=EMAIL, password=PASSWORD)
 
-MODEL_NAME = "test_graphmodule_pt"
+netspresso = NetsPresso(email=EMAIL, password=PASSWORD)
+
+# 1. Declare compressor
+compressor = netspresso.compressor()
+
+# 2. Set variables for compression
 INPUT_SHAPES = [{"batch": 1, "channel": 3, "dimension": [224, 224]}]
 INPUT_MODEL_PATH = "./examples/sample_models/graphmodule.pt"
-OUTPUT_MODEL_PATH = "./outputs/compressed/graphmodule_automatic_compression"
+OUTPUT_DIR = "./outputs/compressed1/pytorch_automatic_compression_1"
 COMPRESSION_RATIO = 0.5
 
+# 3. Run automatic compression
 compressed_model = compressor.automatic_compression(
-    model_name=MODEL_NAME,
     input_shapes=INPUT_SHAPES,
-    input_path=INPUT_MODEL_PATH,
-    output_path=OUTPUT_MODEL_PATH,
+    input_model_path=INPUT_MODEL_PATH,
+    output_dir=OUTPUT_DIR,
     compression_ratio=COMPRESSION_RATIO,
 )

examples/compressor/create_model.py

@@ -1,27 +1,29 @@
-from loguru import logger
-
-from netspresso.compressor import Compressor, Task, Framework
+from netspresso import NetsPresso
+from netspresso.enums import Framework
 
 
 EMAIL = "YOUR_EMAIL"
 PASSWORD = "YOUR_PASSWORD"
-compressor = Compressor(email=EMAIL, password=PASSWORD)
 
-UPLOAD_MODEL_NAME = "test_h5"
-TASK = Task.IMAGE_CLASSIFICATION
+netspresso = NetsPresso(email=EMAIL, password=PASSWORD)
+
+# 1. Declare compressor
+compressor = netspresso.compressor()
+
+# 2. Set variables for upload
+UPLOAD_MODEL_NAME = "sample_h5"
 FRAMEWORK = Framework.TENSORFLOW_KERAS
-UPLOAD_MODEL_PATH = "./examples/sample_models/mobilenetv1.h5"
+INPUT_MODEL_PATH = "./examples/sample_models/mobilenetv1.h5"
 INPUT_SHAPES = [{"batch": 1, "channel": 3, "dimension": [32, 32]}]
 
-# Upload Model
+# 3. Upload model
 model = compressor.upload_model(
     model_name=UPLOAD_MODEL_NAME,
-    task=TASK,
-    framework=FRAMEWORK,
-    file_path=UPLOAD_MODEL_PATH,
+    input_model_path=INPUT_MODEL_PATH,
     input_shapes=INPUT_SHAPES,
+    framework=FRAMEWORK,
 )
 
-# Get Model
+# 4. Get Model
 model = compressor.get_model(model_id=model.model_id)
-logger.info(f"model id: {model.model_id}")
+print(model)

examples/compressor/manual_compression.py

@@ -1,54 +1,48 @@
-from loguru import logger
-
-from netspresso.compressor import (
-    Compressor,
-    CompressionMethod,
-    Policy,
-    LayerNorm,
-    GroupPolicy,
-    Options,
-)
+from netspresso import NetsPresso
+from netspresso.enums import CompressionMethod, Policy, LayerNorm, GroupPolicy, Options
 
 
 EMAIL = "YOUR_EMAIL"
 PASSWORD = "YOUR_PASSWORD"
-compressor = Compressor(email=EMAIL, password=PASSWORD)
 
-# Upload Model
-UPLOAD_MODEL_NAME = "test_pt"
-UPLOAD_MODEL_PATH = "./examples/sample_models/graphmodule.pt"
+netspresso = NetsPresso(email=EMAIL, password=PASSWORD)
+
+# 1. Declare compressor
+compressor = netspresso.compressor()
+
+# 2. Upload model
+MODEL_NAME = "sample_graphmodule_pt"
+INPUT_MODEL_PATH = "./examples/sample_models/graphmodule.pt"
 INPUT_SHAPES = [{"batch": 1, "channel": 3, "dimension": [224, 224]}]
 model = compressor.upload_model(
-    model_name=UPLOAD_MODEL_NAME,
-    file_path=UPLOAD_MODEL_PATH,
+    model_name=MODEL_NAME,
+    input_model_path=INPUT_MODEL_PATH,
     input_shapes=INPUT_SHAPES,
 )
 
-# Select Compression Method
+# 3. Select compression method
 COMPRESSION_METHOD = CompressionMethod.PR_L2
 OPTIONS = Options(
     policy=Policy.AVERAGE,
     layer_norm=LayerNorm.STANDARD_SCORE,
     group_policy=GroupPolicy.AVERAGE,
     reshape_channel_axis=-1,
 )
-compression_1 = compressor.select_compression_method(
+compression_info = compressor.select_compression_method(
     model_id=model.model_id,
     compression_method=COMPRESSION_METHOD,
     options=OPTIONS,
 )
-logger.info(f"compression method: {compression_1.compression_method}")
-logger.info(f"available layers: {compression_1.available_layers}")
+print(f"compression method: {compression_info.compression_method}")
+print(f"available layers: {compression_info.available_layers}")
 
-# Set Compression Params
-for available_layer in compression_1.available_layers[:5]:
+# 4. Set params for compression(ratio or rank)
+for available_layer in compression_info.available_layers[:5]:
     available_layer.values = [0.2]
 
-# Compress Model
-COMPRESSED_MODEL_NAME = "test_l2norm"
-OUTPUT_PATH = "./outputs/compressed/graphmodule_manual"
+# 5. Compress model
+OUTPUT_DIR = "./outputs/compressed/graphmodule_manual"
 compressed_model = compressor.compress_model(
-    compression=compression_1,
-    model_name=COMPRESSED_MODEL_NAME,
-    output_path=OUTPUT_PATH,
+    compression=compression_info,
+    output_dir=OUTPUT_DIR,
 )

examples/compressor/recommendation_compression.py

@@ -1,29 +1,29 @@
-from netspresso.compressor import (
-    Compressor,
-    CompressionMethod,
-    RecommendationMethod,
-)
+from netspresso import NetsPresso
+from netspresso.enums import CompressionMethod, RecommendationMethod
 
 
 EMAIL = "YOUR_EMAIL"
 PASSWORD = "YOUR_PASSWORD"
-compressor = Compressor(email=EMAIL, password=PASSWORD)
 
-# Upload Model
-MODEL_NAME = "test_pt"
+netspresso = NetsPresso(email=EMAIL, password=PASSWORD)
+
+# 1. Declare compressor
+compressor = netspresso.compressor()
+
+# 2. Set variables for compression
 INPUT_MODEL_PATH = "./examples/sample_models/graphmodule.pt"
-OUTPUT_MODEL_PATH = "./outputs/compressed/graphmodule_recommend"
+OUTPUT_DIR = "./outputs/compressed/graphmodule_recommend"
 INPUT_SHAPES = [{"batch": 1, "channel": 3, "dimension": [224, 224]}]
 COMPRESSION_METHOD = CompressionMethod.PR_L2
 RECOMMENDATION_METHOD = RecommendationMethod.SLAMP
 RECOMMENDATION_RATIO = 0.5
 
+# 3. Run recommendation compression
 compressed_model = compressor.recommendation_compression(
-    model_name=MODEL_NAME,
     compression_method=COMPRESSION_METHOD,
     recommendation_method=RECOMMENDATION_METHOD,
     recommendation_ratio=RECOMMENDATION_RATIO,
-    input_path=INPUT_MODEL_PATH,
-    output_path=OUTPUT_MODEL_PATH,
+    input_model_path=INPUT_MODEL_PATH,
+    output_dir=OUTPUT_DIR,
     input_shapes=INPUT_SHAPES,
 )

examples/converter & benchmarker/keras2tflite.py

@@ -0,0 +1,54 @@
+from netspresso import NetsPresso
+from netspresso.enums import Framework, DeviceName
+
+
+###
+# Available target frameworks for conversion with keras models
+#
+# Framework.TENSORFLOW_LITE
+#
+
+###
+# Available devices for Framework.TENSORFLOW_LITE (target_framework)
+#
+# DeviceName.RASPBERRY_PI_4B
+# DeviceName.RASPBERRY_PI_3B_PLUS
+# DeviceName.RASPBERRY_PI_ZERO_W
+# DeviceName.RASPBERRY_PI_ZERO_2W
+#
+
+EMAIL = "YOUR_EMAIL"
+PASSWORD = "YOUR_PASSWORD"
+
+netspresso = NetsPresso(email=EMAIL, password=PASSWORD)
+
+# 1. Declare converter
+converter = netspresso.converter()
+
+# 2. Set variables for convert
+INPUT_MODEL_PATH = "./examples/sample_models/mobilenetv1.h5"
+OUTPUT_DIR = "./outputs/converted/TFLITE_RASPBERRY_PI_4B"
+TARGET_FRAMEWORK = Framework.TENSORFLOW_LITE
+TARGET_DEVICE_NAME = DeviceName.RASPBERRY_PI_4B
+
+# 3. Run convert
+conversion_task = converter.convert_model(
+    input_model_path=INPUT_MODEL_PATH,
+    output_dir=OUTPUT_DIR,
+    target_framework=TARGET_FRAMEWORK,
+    target_device_name=TARGET_DEVICE_NAME,
+)
+print(conversion_task)
+
+# 4. Declare benchmarker
+benchmarker = netspresso.benchmarker()
+
+# 5. Set variables for benchmark
+CONVERTED_MODEL_PATH = "./outputs/converted/TFLITE_RASPBERRY_PI_4B/TFLITE_RASPBERRY_PI_4B.tflite"
+
+# 6. Run benchmark
+benchmark_task = benchmarker.benchmark_model(
+    input_model_path=CONVERTED_MODEL_PATH,
+    target_device_name=TARGET_DEVICE_NAME,
+)
+print(f"model inference latency: {benchmark_task['result']['latency']} ms")

examples/converter & benchmarker/keras2tflite_int8.py

@@ -0,0 +1,59 @@
+from netspresso import NetsPresso
+from netspresso.enums import Framework, DeviceName, DataType
+
+
+# Available target frameworks for conversion with keras models
+#
+# Framework.TENSORFLOW_LITE
+#
+
+# Available devices for Framework.TENSORFLOW_LITE and DataType.INT8 (target_framework)
+#
+# DeviceName.RASPBERRY_PI_4B
+# DeviceName.RASPBERRY_PI_3B_PLUS
+# DeviceName.RASPBERRY_PI_ZERO_W
+# DeviceName.RASPBERRY_PI_ZERO_2W
+# DeviceName.ALIF_ENSEMBLE_E7_DEVKIT_GEN2
+# DeviceName.RENESAS_RA8D1
+#
+
+EMAIL = "YOUR_EMAIL"
+PASSWORD = "YOUR_PASSWORD"
+
+netspresso = NetsPresso(email=EMAIL, password=PASSWORD)
+
+# 1. Declare converter
+converter = netspresso.converter()
+
+# 2. Set variables for convert
+INPUT_MODEL_PATH = "./examples/sample_models/mobilenetv1_cifar100_automatic.h5"
+OUTPUT_DIR = "./outputs/converted/TFLITE_INT8_RENESAS_RA8D1"
+TARGET_FRAMEWORK = Framework.TENSORFLOW_LITE
+TARGET_DEVICE_NAME = DeviceName.RENESAS_RA8D1
+TARGET_DATA_TYPE = DataType.INT8
+DATASET_PATH = "./examples/sample_datasets/20x32x32x3.npy"
+
+# 3. Run convert
+conversion_task = converter.convert_model(
+    input_model_path=INPUT_MODEL_PATH,
+    output_dir=OUTPUT_DIR,
+    target_framework=TARGET_FRAMEWORK,
+    target_data_type=TARGET_DATA_TYPE,
+    target_device_name=TARGET_DEVICE_NAME,
+    dataset_path=DATASET_PATH,
+)
+print(conversion_task)
+
+# 4. Declare benchmarker
+benchmarker = netspresso.benchmarker()
+
+# 5. Set variables for benchmark
+CONVERTED_MODEL_PATH = "./outputs/converted/TFLITE_INT8_RENESAS_RA8D1/TFLITE_INT8_RENESAS_RA8D1.tflite"
+
+# 6. Run benchmark
+benchmark_task = benchmarker.benchmark_model(
+    input_model_path=CONVERTED_MODEL_PATH,
+    target_data_type=TARGET_DATA_TYPE,
+    target_device_name=TARGET_DEVICE_NAME,
+)
+print(f"model inference latency: {benchmark_task['result']['latency']} ms")