optimization/quantization added into 500 series (#2197)

Co-authored-by: Samet Akcay <samet.akcay@intel.com>
openvinotoolkit · Aug 22, 2024 · 775f5a8 · 775f5a8
1 parent cfd3d8e
commit 775f5a8
Showing 1 changed file with 39 additions and 0 deletions.
diff --git a/notebooks/500_use_cases/501_dobot/501a_training_a_model_with_cubes_from_a_robotic_arm.ipynb b/notebooks/500_use_cases/501_dobot/501a_training_a_model_with_cubes_from_a_robotic_arm.ipynb
@@ -437,6 +437,45 @@
     ")"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "For optimization and quantization process, we are using a seamless integration with [NNCF Library](https://github.com/openvinotoolkit/nncf) in the backend of Anomalib. Select one of the following options for optimization or quantization. Replace the openvino_model_path line above in order to export the optimized/quantized model:\n",
+    "\n",
+    "```\n",
+    "# Exporting optimized/quantized models\n",
+    "\n",
+    "# Post Training Quantization\n",
+    "openvino_model_path = engine.export(\n",
+    "    model, \n",
+    "    ExportType.OPENVINO, \n",
+    "    str(Path.cwd()) + \"_optimized\", \n",
+    "    compression_type=CompressionType.INT8_PTQ, \n",
+    "    datamodule=datamodule\n",
+    "    )\n",
+    "\n",
+    "# Accuracy-Control Quantization\n",
+    "openvino_model_path=engine.export(\n",
+    "    model, \n",
+    "    ExportType.OPENVINO, \n",
+    "    str(Path.cwd()) + \"_optimized\", \n",
+    "    compression_type=CompressionType.INT8_ACQ, \n",
+    "    datamodule=datamodule, \n",
+    "    metric=\"F1Score\"\n",
+    "    )\n",
+    "\n",
+    "# Weight Compression\n",
+    "openvino_model_path=engine.export(\n",
+    "    model, \n",
+    "    ExportType.OPENVINO, \n",
+    "    str(Path.cwd()) + \"_WEIGHTS\", \n",
+    "    compression_type=CompressionType.FP16, \n",
+    "    datamodule=datamodule\n",
+    "    )\n",
+    "```"
+   ]
+  },
   {
    "attachments": {},
    "cell_type": "markdown",