Spaces:
Runtime error
Runtime error
File size: 53,158 Bytes
db5855f |
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 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 |
{
"cells": [
{
"attachments": {},
"cell_type": "markdown",
"id": "8e5ea381-9420-4ce0-8dea-7bbb4f3f3daa",
"metadata": {},
"source": [
"# Hello NPU\n",
"\n",
"## Working with NPU in OpenVINO™\n",
"\n",
"#### Table of contents:\n",
"- [Introduction](#Introduction)\n",
" - [Install required packages](#Install-required-packages)\n",
"- [Checking NPU with Query Device](#Checking-NPU-with-Query-Device)\n",
" - [List the NPU with core.available_devices](#List-the-NPU-with-core.available_devices)\n",
" - [Check Properties with core.get_property](#Check-Properties-with-core.get_property)\n",
" - [Brief Descriptions of Key Properties](#Brief-Descriptions-of-Key-Properties)\n",
"- [Compiling a Model on NPU](#Compiling-a-Model-on-NPU)\n",
" - [Download and Convert a Model](#Download-and-Convert-a-Model)\n",
" - [Download the Model](#Download-the-Model)\n",
" - [Convert the Model to OpenVINO IR format](#Convert-the-Model-to-OpenVINO-IR-format)\n",
" - [Compile with Default Configuration](#Compile-with-Default-Configuration)\n",
" - [Reduce Compile Time through Model Caching](#Reduce-Compile-Time-through-Model-Caching)\n",
" - [UMD Model Caching](#UMD-Model-Caching)\n",
" - [OpenVINO Model Caching](#OpenVINO-Model-Caching)\n",
" - [Throughput and Latency Performance Hints](#Throughput-and-Latency-Performance-Hints)\n",
"- [Performance Comparison with benchmark_app](#Performance-Comparison-with-benchmark_app)\n",
" - [NPU vs CPU with Latency Hint](#NPU-vs-CPU-with-Latency-Hint)\n",
" - [Effects of UMD Model Caching](#Effects-of-UMD-Model-Caching)\n",
" - [NPU vs CPU with Throughput Hint](#NPU-vs-CPU-with-Throughput-Hint)\n",
"- [Limitations](#Limitations)\n",
"- [Conclusion](#Conclusion)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "b6a83166-b033-4b20-b223-5eec6b84f46b",
"metadata": {},
"source": [
"This tutorial provides a high-level overview of working with the NPU device **Intel(R) AI Boost** (introduced with the Intel® Core™ Ultra generation of CPUs) in OpenVINO. It explains some of the key properties of the NPU and shows how to compile a model on NPU with performance hints.\n",
"\n",
"This tutorial also shows example commands for benchmark_app that can be run to compare NPU performance with CPU in different configurations."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "45805698-72df-4197-b6e5-d57fdc3a365e",
"metadata": {},
"source": [
"## Introduction\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "bdfc9e1f-f463-4ffe-820b-c924857fb066",
"metadata": {},
"source": [
"The Neural Processing Unit (NPU) is a low power hardware solution which enables you to offload certain neural network computation tasks from other devices, for more streamlined resource management.\n",
"\n",
"Note that the NPU plugin is included in PIP installation of OpenVINO™ and you need to [install a proper NPU driver](https://docs.openvino.ai/2024/get-started/configurations/configurations-intel-npu.html) to use it successfully.\n",
"\n",
"**Supported Platforms**: \n",
" Host: Intel® Core™ Ultra \n",
" NPU device: NPU 3720 \n",
" OS: Ubuntu 22.04 (with Linux Kernel 6.6+), MS Windows 11 (both 64-bit)\n",
"\n",
"To learn more about the NPU Device, see the [page](https://docs.openvino.ai/2024/openvino-workflow/running-inference/inference-devices-and-modes/npu-device.html).\n",
" "
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "3d568b8f-5f53-42f6-82b9-ee29412b4f96",
"metadata": {},
"source": [
"### Install required packages\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "69c5b00e-0ec4-4c57-b9c3-1a25e2c264c2",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Note: you may need to restart the kernel to use updated packages.\n"
]
}
],
"source": [
"%pip install -q \"openvino>=2024.1.0\" torch torchvision --extra-index-url https://download.pytorch.org/whl/cpu"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "e19f77db-f27a-493c-b389-af520eac5422",
"metadata": {},
"source": [
"## Checking NPU with Query Device\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "c632f40b-1a59-4c9e-b8e0-6eec3ca81703",
"metadata": {},
"source": [
"In this section, we will see how to list the available NPU and check its properties. Some of the key properties will be defined."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "731e32fb-9a74-4e76-8f77-62f16b0d49a8",
"metadata": {},
"source": [
"### List the NPU with core.available_devices\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "84eab2e7-36f6-45f5-84d5-7a946afc7534",
"metadata": {},
"source": [
"OpenVINO Runtime provides the ```available_devices``` method for checking which devices are available for inference. The following code will output a list a compatible OpenVINO devices, in which Intel NPU should appear (ensure that the driver is installed successfully). "
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "2f930769-b305-4f68-8cce-a86258f80af2",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"['CPU', 'GPU', 'NPU']"
]
},
"execution_count": 2,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"import openvino as ov\n",
"\n",
"core = ov.Core()\n",
"core.available_devices"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "87cb433f-7c7e-4e0f-b0b3-31be4eabc8d1",
"metadata": {},
"source": [
"### Check Properties with core.get_property\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "1b97c7f8-f697-4df6-8bfd-ec2fbfb8273d",
"metadata": {},
"source": [
"To get information about the NPU, we can use device properties. In OpenVINO, devices have properties that describe their characteristics and configurations. Each property has a name and associated value that can be queried with the ```get_property``` method."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "c60fa7b1-0b43-4a92-ac6f-d4e9d96c88fd",
"metadata": {},
"source": [
"To get the value of a property, such as the device name, we can use the ```get_property``` method as follows:"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "861af905-a574-4c74-a8a9-e7735c9df43a",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"'Intel(R) AI Boost'"
]
},
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"device = \"NPU\"\n",
"\n",
"core.get_property(device, \"FULL_DEVICE_NAME\")"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "70889c34-74f8-4a7a-b23d-166311c7c02d",
"metadata": {},
"source": [
"Each device also has a specific property called ```SUPPORTED_PROPERTIES```, that enables viewing all the available properties in the device. We can check the value for each property by simply looping through the dictionary returned by ```core.get_property(\"NPU\", \"SUPPORTED_PROPERTIES\")``` and then querying for that property."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "3560b0f4-e105-4004-aefe-d685667925c8",
"metadata": {},
"outputs": [],
"source": [
"print(f\"{device} SUPPORTED_PROPERTIES:\\n\")\n",
"supported_properties = core.get_property(device, \"SUPPORTED_PROPERTIES\")\n",
"indent = len(max(supported_properties, key=len))\n",
"\n",
"for property_key in supported_properties:\n",
" if property_key not in (\"SUPPORTED_METRICS\", \"SUPPORTED_CONFIG_KEYS\", \"SUPPORTED_PROPERTIES\"):\n",
" try:\n",
" property_val = core.get_property(device, property_key)\n",
" except TypeError:\n",
" property_val = \"UNSUPPORTED TYPE\"\n",
" print(f\"{property_key:<{indent}}: {property_val}\")"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "890be759-744a-49f1-9261-81feee9e09c4",
"metadata": {},
"source": [
"### Brief Descriptions of Key Properties\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "4a5f17fe-1f06-4026-9668-196853d55278",
"metadata": {},
"source": [
"Each device has several properties as seen in the last command. Some of the key properties are:\n",
"- `FULL_DEVICE_NAME` - The product name of the NPU.\n",
"- `PERFORMANCE_HINT` - A high-level way to tune the device for a specific performance metric, such as latency or throughput, without worrying about device-specific settings.\n",
"- `CACHE_DIR` - The directory where the OpenVINO model cache data is stored to speed up the compilation time.\n",
"- `OPTIMIZATION_CAPABILITIES` - The model data types (INT8, FP16, FP32, etc) that are supported by this NPU.\n",
"\n",
"To learn more about devices and properties, see the [Query Device Properties](https://docs.openvino.ai/2024/openvino-workflow/running-inference/inference-devices-and-modes/query-device-properties.html) page."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "1f95f2fa-d804-480e-831e-b118f637e646",
"metadata": {},
"source": [
"## Compiling a Model on NPU\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "77317cda-9e17-45bc-8122-af6f4260f3ce",
"metadata": {},
"source": [
"Now, we know the NPU present in the system and we have checked its properties. We can easily use it for compiling and running models with OpenVINO NPU plugin."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "1ac8055c-b45c-478d-bc75-ec68b8b17ff2",
"metadata": {},
"source": [
"### Download and Convert a Model\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "6df3a6aa-f35f-482f-bb12-1d2bee2575fd",
"metadata": {},
"source": [
"This tutorial uses the `resnet50` model. The `resnet50` model is used for image classification tasks. The model was trained on [ImageNet](https://www.image-net.org/index.php) dataset which contains over a million images categorized into 1000 classes. To read more about resnet50, see the [paper](https://ieeexplore.ieee.org/document/7780459)."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "1998296a-eb8e-427d-b28f-1296916c8b11",
"metadata": {},
"source": [
"#### Download the Model\n",
"[back to top ⬆️](#Table-of-contents:)\n",
"\n",
"Fetch [ResNet50 CV](https://pytorch.org/vision/stable/models/generated/torchvision.models.resnet50.html#torchvision.models.ResNet50_Weights) Classification model from torchvision."
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "f6dd041c-6927-4f3a-a779-21cb90565811",
"metadata": {},
"outputs": [],
"source": [
"from pathlib import Path\n",
"\n",
"# create a directory for resnet model file\n",
"MODEL_DIRECTORY_PATH = Path(\"model\")\n",
"MODEL_DIRECTORY_PATH.mkdir(exist_ok=True)\n",
"\n",
"model_name = \"resnet50\""
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "50840dd5-4792-45b1-bbea-a66ac581b9b8",
"metadata": {},
"outputs": [],
"source": [
"from torchvision.models import resnet50, ResNet50_Weights\n",
"\n",
"# create model object\n",
"pytorch_model = resnet50(weights=ResNet50_Weights.DEFAULT)\n",
"\n",
"# switch model from training to inference mode\n",
"pytorch_model.eval();"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "6133b715-c1b4-4a2a-bfd8-38da81e25cdb",
"metadata": {},
"source": [
"#### Convert the Model to OpenVINO IR format\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "c662a851-1b51-4a13-b725-6533fec8c8cc",
"metadata": {},
"source": [
"To convert this Pytorch model to OpenVINO IR with `FP16` precision, use model conversion API. The models are saved to the `model/ir_model/` directory. For more details about model conversion, see this [page](https://docs.openvino.ai/2024/openvino-workflow/model-preparation.html)."
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "5cea2106-3274-4b3f-9fad-882f27dd9a0b",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Read IR model from model\\ir_model\\resnet50_fp16.xml\n"
]
}
],
"source": [
"precision = \"FP16\"\n",
"\n",
"model_path = MODEL_DIRECTORY_PATH / \"ir_model\" / f\"{model_name}_{precision.lower()}.xml\"\n",
"\n",
"model = None\n",
"if not model_path.exists():\n",
" model = ov.convert_model(pytorch_model, input=[[1, 3, 224, 224]])\n",
" ov.save_model(model, model_path, compress_to_fp16=(precision == \"FP16\"))\n",
" print(\"IR model saved to {}\".format(model_path))\n",
"else:\n",
" print(\"Read IR model from {}\".format(model_path))\n",
" model = core.read_model(model_path)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "2a3af027-41bf-4281-845e-41b445e99f3d",
"metadata": {},
"source": [
"**Note:** NPU also supports `INT8` quantized models."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "c665ad4f-2c2c-4d1a-b429-92279c5b6629",
"metadata": {},
"source": [
"### Compile with Default Configuration\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "1bbbdd98-ffd3-4568-b6b3-eb1ce8528c39",
"metadata": {},
"source": [
"When the model is ready, first we need to read it, using the `read_model` method. Then, we can use the `compile_model` method and specify the name of the device we want to compile the model on, in this case, \"NPU\"."
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "95d7a069-60d7-40e0-8dd9-82f6727cc9f5",
"metadata": {},
"outputs": [],
"source": [
"compiled_model = core.compile_model(model, device)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "f4ff6268-618b-44bf-8af3-70b291c79dca",
"metadata": {},
"source": [
"### Reduce Compile Time through Model Caching\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "4cf2ae0e-d652-4e1e-b461-d382eee0257e",
"metadata": {},
"source": [
"Depending on the model used, device-specific optimizations and network compilations can cause the compile step to be time-consuming, especially with larger models, which may lead to bad user experience in the application. To solve this **Model Caching** can be used."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "3388175a-9f43-412e-be84-6a2a3dad9799",
"metadata": {},
"source": [
"Model Caching helps reduce application startup delays by exporting and reusing the compiled model automatically. The following two compilation-related metrics are crucial in this area:\n",
"\n",
"- **First-Ever Inference Latency (FEIL)**: \n",
" Measures all steps required to compile and execute a model on the device for the first time. It includes model compilation time, the time required to load and initialize the model on the device and the first inference execution.\n",
"- **First Inference Latency (FIL)**: \n",
" Measures the time required to load and initialize the pre-compiled model on the device and the first inference execution."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "661175bc-6497-4329-89c9-c18df52d5b21",
"metadata": {},
"source": [
"In NPU, UMD model caching is a solution enabled by default by the driver. It improves time to first inference (FIL) by storing the model in the cache after compilation (included in FEIL). Learn more about UMD Caching [here](https://docs.openvino.ai/2024/openvino-workflow/running-inference/inference-devices-and-modes/npu-device.html#umd-dynamic-model-caching). Due to this caching, it takes lesser time to load the model after first compilation.\n",
"\n",
"You can also use OpenVINO Model Caching, which is a common mechanism for all OpenVINO device plugins and can be enabled by setting the `cache_dir` property. \n",
"By enabling OpenVINO Model Caching, the UMD caching is automatically bypassed by the NPU plugin, which means the model will only be stored in the OpenVINO cache after compilation. When a cache hit occurs for subsequent compilation requests, the plugin will import the model instead of recompiling it."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "f21103bc-82ff-4f8a-8ccc-bb8a30219e0f",
"metadata": {},
"source": [
"#### UMD Model Caching\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "ad548de7-b591-454e-b049-7ce8c50a2975",
"metadata": {},
"source": [
"To see how UMD caching see the following example:"
]
},
{
"cell_type": "code",
"execution_count": 30,
"id": "8719d2ec-2866-4f3b-ba62-098f9190ce6c",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"UMD Caching (first time) - compile time: 3.2854952812194824s\n"
]
}
],
"source": [
"import time\n",
"from pathlib import Path\n",
"\n",
"start = time.time()\n",
"core = ov.Core()\n",
"\n",
"# Compile the model as before\n",
"model = core.read_model(model=model_path)\n",
"compiled_model = core.compile_model(model, device)\n",
"print(f\"UMD Caching (first time) - compile time: {time.time() - start}s\")"
]
},
{
"cell_type": "code",
"execution_count": 31,
"id": "3cdfeaa5-8f2e-4e94-b45f-dd23b14e0408",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"UMD Caching - compile time: 2.269814968109131s\n"
]
}
],
"source": [
"start = time.time()\n",
"core = ov.Core()\n",
"\n",
"# Compile the model once again to see UMD Caching\n",
"model = core.read_model(model=model_path)\n",
"compiled_model = core.compile_model(model, device)\n",
"print(f\"UMD Caching - compile time: {time.time() - start}s\")"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "b4be65ac-e2fc-4ac1-ab73-9f72025992a5",
"metadata": {},
"source": [
"#### OpenVINO Model Caching\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "335fa1fa-5acf-4ee0-94d7-0f56c5546aed",
"metadata": {},
"source": [
"To get an idea of OpenVINO model caching, we can use the OpenVINO cache as follow"
]
},
{
"cell_type": "code",
"execution_count": 19,
"id": "1e5683f1-514d-4c72-b4ae-683849ca7be9",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Cache enabled (first time) - compile time: 0.6362860202789307s\n",
"Cache enabled (second time) - compile time: 0.3032548427581787s\n"
]
}
],
"source": [
"# Create cache folder\n",
"cache_folder = Path(\"cache\")\n",
"cache_folder.mkdir(exist_ok=True)\n",
"\n",
"start = time.time()\n",
"core = ov.Core()\n",
"\n",
"# Set cache folder\n",
"core.set_property({\"CACHE_DIR\": cache_folder})\n",
"\n",
"# Compile the model\n",
"model = core.read_model(model=model_path)\n",
"compiled_model = core.compile_model(model, device)\n",
"print(f\"Cache enabled (first time) - compile time: {time.time() - start}s\")\n",
"\n",
"start = time.time()\n",
"core = ov.Core()\n",
"\n",
"# Set cache folder\n",
"core.set_property({\"CACHE_DIR\": cache_folder})\n",
"\n",
"# Compile the model as before\n",
"model = core.read_model(model=model_path)\n",
"compiled_model = core.compile_model(model, device)\n",
"print(f\"Cache enabled (second time) - compile time: {time.time() - start}s\")"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "f05a870a-686a-44f2-9b2f-1e4f9ccf2447",
"metadata": {},
"source": [
"And when the OpenVINO cache is disabled:"
]
},
{
"cell_type": "code",
"execution_count": 20,
"id": "39f336b6-0237-4cec-b8af-fbf371351cce",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Cache disabled - compile time: 3.0127954483032227s\n"
]
}
],
"source": [
"start = time.time()\n",
"core = ov.Core()\n",
"model = core.read_model(model=model_path)\n",
"compiled_model = core.compile_model(model, device)\n",
"print(f\"Cache disabled - compile time: {time.time() - start}s\")"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "6b55b71b-f6f6-4619-aed1-d216dc83ab9c",
"metadata": {},
"source": [
"The actual time improvements will depend on the environment as well as the model being used but it is definitely something to consider when optimizing an application. To read more about this, see the [Model Caching docs](https://docs.openvino.ai/2024/openvino-workflow/running-inference/optimize-inference/optimizing-latency/model-caching-overview.html)."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "9240250b-15ac-46b4-b1ba-055f04a7ed15",
"metadata": {},
"source": [
"### Throughput and Latency Performance Hints\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "1610f4ba-748c-44d1-acae-52f046cc7697",
"metadata": {},
"source": [
"To simplify device and pipeline configuration, OpenVINO provides high-level performance hints that automatically set the batch size and number of parallel threads for inference. The \"LATENCY\" performance hint optimizes for fast inference times while the \"THROUGHPUT\" performance hint optimizes for high overall bandwidth or FPS."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "1cccd1b5-4d5a-41f3-8d8a-4ee0bc235a9e",
"metadata": {},
"source": [
"To use the \"LATENCY\" performance hint, add `{\"PERFORMANCE_HINT\": \"LATENCY\"}` when compiling the model as shown below. For NPU, this automatically minimizes the batch size and number of parallel streams such that all of the compute resources can focus on completing a single inference as fast as possible."
]
},
{
"cell_type": "code",
"execution_count": 21,
"id": "0773486e-8294-455c-8d13-c803c4c68961",
"metadata": {},
"outputs": [],
"source": [
"compiled_model = core.compile_model(model, device, {\"PERFORMANCE_HINT\": \"LATENCY\"})"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "7ca1f3d8-202c-4a98-85bc-b66110120dfb",
"metadata": {},
"source": [
"To use the \"THROUGHPUT\" performance hint, add `{\"PERFORMANCE_HINT\": \"THROUGHPUT\"}` when compiling the model. For NPUs, this creates multiple processing streams to efficiently utilize all the execution cores and optimizes the batch size to fill the available memory."
]
},
{
"cell_type": "code",
"execution_count": 22,
"id": "162dc563-c80e-4e4b-ad77-20d80715a19c",
"metadata": {},
"outputs": [],
"source": [
"compiled_model = core.compile_model(model, device, {\"PERFORMANCE_HINT\": \"THROUGHPUT\"})"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "0328676a-f569-4a4d-bda0-3fb47b97a420",
"metadata": {},
"source": [
"## Performance Comparison with benchmark_app\n",
"[back to top ⬆️](#Table-of-contents:)\n"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "41396503-d589-4b98-be91-e2277d6c7808",
"metadata": {},
"source": [
"Given all the different options available when compiling a model, it may be difficult to know which settings work best for a certain application. Thankfully, OpenVINO provides `benchmark_app` - a performance benchmarking tool."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "864304b8-e129-4d14-adcc-efe3ad350456",
"metadata": {},
"source": [
"The basic syntax of `benchmark_app` is as follows:\n",
"\n",
"`\n",
"benchmark_app -m PATH_TO_MODEL -d TARGET_DEVICE -hint {throughput,cumulative_throughput,latency,none}\n",
"`\n",
"\n",
"where `TARGET_DEVICE` is any device shown by the `available_devices` method as well as the MULTI and AUTO devices we saw previously, and the value of hint should be one of the values between brackets.\n",
"\n",
"Note that benchmark_app only requires the model path to run but both device and hint arguments will be useful to us. For more advanced usages, the tool itself has other options that can be checked by running `benchmark_app -h` or reading the [docs](https://docs.openvino.ai/2024/learn-openvino/openvino-samples/benchmark-tool.html). The following example shows us to benchmark a simple model, using a NPU with latency focus:"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "7915a0b1-f0a9-41b5-b38b-3cbf97a69a7d",
"metadata": {},
"source": [
"`benchmark_app -m {model_path} -d NPU -hint latency`"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "86af2b7b-6723-403a-80f9-0ee59c9429b2",
"metadata": {},
"source": [
"For completeness, let us list here some of the comparisons we may want to do by varying the device and hint used. Note that the actual performance may depend on the hardware used. Generally, we should expect NPU to be better than CPU. \n",
"Please refer to the `benchmark_app` log entries under `[Step 11/11] Dumping statistics report` to observe the differences in latency and throughput between the CPU and NPU.."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "de8a7194-60d9-4b74-ac8b-f663241c60f4",
"metadata": {},
"source": [
"#### NPU vs CPU with Latency Hint\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"cell_type": "code",
"execution_count": 23,
"id": "183bc1d4-b0a6-4b3a-ae8d-7fcb21d596ea",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[Step 1/11] Parsing and validating input arguments\n",
"[ INFO ] Parsing input parameters\n",
"[Step 2/11] Loading OpenVINO Runtime\n",
"[ INFO ] OpenVINO:\n",
"[ INFO ] Build ................................. 2024.1.0-14992-621b025bef4\n",
"[ INFO ] \n",
"[ INFO ] Device info:\n",
"[ INFO ] CPU\n",
"[ INFO ] Build ................................. 2024.1.0-14992-621b025bef4\n",
"[ INFO ] \n",
"[ INFO ] \n",
"[Step 3/11] Setting device configuration\n",
"[Step 4/11] Reading model files\n",
"[ INFO ] Loading model files\n",
"[ INFO ] Read model took 14.00 ms\n",
"[ INFO ] Original model I/O parameters:\n",
"[ INFO ] Model inputs:\n",
"[ INFO ] x (node: x) : f32 / [...] / [1,3,224,224]\n",
"[ INFO ] Model outputs:\n",
"[ INFO ] x.45 (node: aten::linear/Add) : f32 / [...] / [1,1000]\n",
"[Step 5/11] Resizing model to match image sizes and given batch\n",
"[ INFO ] Model batch size: 1\n",
"[Step 6/11] Configuring input of the model\n",
"[ INFO ] Model inputs:\n",
"[ INFO ] x (node: x) : u8 / [N,C,H,W] / [1,3,224,224]\n",
"[ INFO ] Model outputs:\n",
"[ INFO ] x.45 (node: aten::linear/Add) : f32 / [...] / [1,1000]\n",
"[Step 7/11] Loading the model to the device\n",
"[ INFO ] Compile model took 143.22 ms\n",
"[Step 8/11] Querying optimal runtime parameters\n",
"[ INFO ] Model:\n",
"[ INFO ] NETWORK_NAME: Model2\n",
"[ INFO ] OPTIMAL_NUMBER_OF_INFER_REQUESTS: 1\n",
"[ INFO ] NUM_STREAMS: 1\n",
"[ INFO ] AFFINITY: Affinity.HYBRID_AWARE\n",
"[ INFO ] INFERENCE_NUM_THREADS: 12\n",
"[ INFO ] PERF_COUNT: NO\n",
"[ INFO ] INFERENCE_PRECISION_HINT: <Type: 'float32'>\n",
"[ INFO ] PERFORMANCE_HINT: LATENCY\n",
"[ INFO ] EXECUTION_MODE_HINT: ExecutionMode.PERFORMANCE\n",
"[ INFO ] PERFORMANCE_HINT_NUM_REQUESTS: 0\n",
"[ INFO ] ENABLE_CPU_PINNING: False\n",
"[ INFO ] SCHEDULING_CORE_TYPE: SchedulingCoreType.ANY_CORE\n",
"[ INFO ] MODEL_DISTRIBUTION_POLICY: set()\n",
"[ INFO ] ENABLE_HYPER_THREADING: False\n",
"[ INFO ] EXECUTION_DEVICES: ['CPU']\n",
"[ INFO ] CPU_DENORMALS_OPTIMIZATION: False\n",
"[ INFO ] LOG_LEVEL: Level.NO\n",
"[ INFO ] CPU_SPARSE_WEIGHTS_DECOMPRESSION_RATE: 1.0\n",
"[ INFO ] DYNAMIC_QUANTIZATION_GROUP_SIZE: 0\n",
"[ INFO ] KV_CACHE_PRECISION: <Type: 'float16'>\n",
"[Step 9/11] Creating infer requests and preparing input tensors\n",
"[ WARNING ] No input files were given for input 'x'!. This input will be filled with random values!\n",
"[ INFO ] Fill input 'x' with random values \n",
"[Step 10/11] Measuring performance (Start inference asynchronously, 1 inference requests, limits: 60000 ms duration)\n",
"[ INFO ] Benchmarking in inference only mode (inputs filling are not included in measurement loop).\n",
"[ INFO ] First inference took 28.95 ms\n",
"[Step 11/11] Dumping statistics report\n",
"[ INFO ] Execution Devices:['CPU']\n",
"[ INFO ] Count: 1612 iterations\n",
"[ INFO ] Duration: 60039.72 ms\n",
"[ INFO ] Latency:\n",
"[ INFO ] Median: 39.99 ms\n",
"[ INFO ] Average: 37.13 ms\n",
"[ INFO ] Min: 19.13 ms\n",
"[ INFO ] Max: 71.94 ms\n",
"[ INFO ] Throughput: 26.85 FPS\n"
]
}
],
"source": [
"!benchmark_app -m {model_path} -d CPU -hint latency"
]
},
{
"cell_type": "code",
"execution_count": 24,
"id": "d5aa1702-9118-4c62-bba3-ca6ca1cf0961",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[Step 1/11] Parsing and validating input arguments\n",
"[ INFO ] Parsing input parameters\n",
"[Step 2/11] Loading OpenVINO Runtime\n",
"[ INFO ] OpenVINO:\n",
"[ INFO ] Build ................................. 2024.1.0-14992-621b025bef4\n",
"[ INFO ] \n",
"[ INFO ] Device info:\n",
"[ INFO ] NPU\n",
"[ INFO ] Build ................................. 2024.1.0-14992-621b025bef4\n",
"[ INFO ] \n",
"[ INFO ] \n",
"[Step 3/11] Setting device configuration\n",
"[Step 4/11] Reading model files\n",
"[ INFO ] Loading model files\n",
"[ INFO ] Read model took 11.51 ms\n",
"[ INFO ] Original model I/O parameters:\n",
"[ INFO ] Model inputs:\n",
"[ INFO ] x (node: x) : f32 / [...] / [1,3,224,224]\n",
"[ INFO ] Model outputs:\n",
"[ INFO ] x.45 (node: aten::linear/Add) : f32 / [...] / [1,1000]\n",
"[Step 5/11] Resizing model to match image sizes and given batch\n",
"[ INFO ] Model batch size: 1\n",
"[Step 6/11] Configuring input of the model\n",
"[ INFO ] Model inputs:\n",
"[ INFO ] x (node: x) : u8 / [N,C,H,W] / [1,3,224,224]\n",
"[ INFO ] Model outputs:\n",
"[ INFO ] x.45 (node: aten::linear/Add) : f32 / [...] / [1,1000]\n",
"[Step 7/11] Loading the model to the device\n",
"[ INFO ] Compile model took 2302.40 ms\n",
"[Step 8/11] Querying optimal runtime parameters\n",
"[ INFO ] Model:\n",
"[ INFO ] DEVICE_ID: \n",
"[ INFO ] ENABLE_CPU_PINNING: False\n",
"[ INFO ] EXECUTION_DEVICES: NPU.3720\n",
"[ INFO ] INFERENCE_PRECISION_HINT: <Type: 'float16'>\n",
"[ INFO ] INTERNAL_SUPPORTED_PROPERTIES: {'CACHING_PROPERTIES': 'RO'}\n",
"[ INFO ] LOADED_FROM_CACHE: False\n",
"[ INFO ] NETWORK_NAME: \n",
"[ INFO ] OPTIMAL_NUMBER_OF_INFER_REQUESTS: 1\n",
"[ INFO ] PERFORMANCE_HINT: PerformanceMode.LATENCY\n",
"[ INFO ] PERFORMANCE_HINT_NUM_REQUESTS: 1\n",
"[ INFO ] PERF_COUNT: False\n",
"[Step 9/11] Creating infer requests and preparing input tensors\n",
"[ WARNING ] No input files were given for input 'x'!. This input will be filled with random values!\n",
"[ INFO ] Fill input 'x' with random values \n",
"[Step 10/11] Measuring performance (Start inference asynchronously, 1 inference requests, limits: 60000 ms duration)\n",
"[ INFO ] Benchmarking in inference only mode (inputs filling are not included in measurement loop).\n",
"[ INFO ] First inference took 7.94 ms\n",
"[Step 11/11] Dumping statistics report\n",
"[ INFO ] Execution Devices:NPU.3720\n",
"[ INFO ] Count: 17908 iterations\n",
"[ INFO ] Duration: 60004.49 ms\n",
"[ INFO ] Latency:\n",
"[ INFO ] Median: 3.29 ms\n",
"[ INFO ] Average: 3.33 ms\n",
"[ INFO ] Min: 3.21 ms\n",
"[ INFO ] Max: 6.90 ms\n",
"[ INFO ] Throughput: 298.44 FPS\n"
]
}
],
"source": [
"!benchmark_app -m {model_path} -d NPU -hint latency"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "5f9714ec-dd99-4aa8-bfc4-9274df57d7fb",
"metadata": {},
"source": [
"##### Effects of UMD Model Caching\n",
"[back to top ⬆️](#Table-of-contents:)\n",
"\n",
"To see the effects of UMD Model caching, we are going to run the benchmark_app and see the difference in model read time and compilation time:"
]
},
{
"cell_type": "code",
"execution_count": 25,
"id": "8c5c1982-909d-42b2-95f2-ec6f0c7790dc",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[Step 1/11] Parsing and validating input arguments\n",
"[ INFO ] Parsing input parameters\n",
"[Step 2/11] Loading OpenVINO Runtime\n",
"[ INFO ] OpenVINO:\n",
"[ INFO ] Build ................................. 2024.1.0-14992-621b025bef4\n",
"[ INFO ] \n",
"[ INFO ] Device info:\n",
"[ INFO ] NPU\n",
"[ INFO ] Build ................................. 2024.1.0-14992-621b025bef4\n",
"[ INFO ] \n",
"[ INFO ] \n",
"[Step 3/11] Setting device configuration\n",
"[Step 4/11] Reading model files\n",
"[ INFO ] Loading model files\n",
"[ INFO ] Read model took 11.00 ms\n",
"[ INFO ] Original model I/O parameters:\n",
"[ INFO ] Model inputs:\n",
"[ INFO ] x (node: x) : f32 / [...] / [1,3,224,224]\n",
"[ INFO ] Model outputs:\n",
"[ INFO ] x.45 (node: aten::linear/Add) : f32 / [...] / [1,1000]\n",
"[Step 5/11] Resizing model to match image sizes and given batch\n",
"[ INFO ] Model batch size: 1\n",
"[Step 6/11] Configuring input of the model\n",
"[ INFO ] Model inputs:\n",
"[ INFO ] x (node: x) : u8 / [N,C,H,W] / [1,3,224,224]\n",
"[ INFO ] Model outputs:\n",
"[ INFO ] x.45 (node: aten::linear/Add) : f32 / [...] / [1,1000]\n",
"[Step 7/11] Loading the model to the device\n",
"[ INFO ] Compile model took 2157.58 ms\n",
"[Step 8/11] Querying optimal runtime parameters\n",
"[ INFO ] Model:\n",
"[ INFO ] DEVICE_ID: \n",
"[ INFO ] ENABLE_CPU_PINNING: False\n",
"[ INFO ] EXECUTION_DEVICES: NPU.3720\n",
"[ INFO ] INFERENCE_PRECISION_HINT: <Type: 'float16'>\n",
"[ INFO ] INTERNAL_SUPPORTED_PROPERTIES: {'CACHING_PROPERTIES': 'RO'}\n",
"[ INFO ] LOADED_FROM_CACHE: False\n",
"[ INFO ] NETWORK_NAME: \n",
"[ INFO ] OPTIMAL_NUMBER_OF_INFER_REQUESTS: 1\n",
"[ INFO ] PERFORMANCE_HINT: PerformanceMode.LATENCY\n",
"[ INFO ] PERFORMANCE_HINT_NUM_REQUESTS: 1\n",
"[ INFO ] PERF_COUNT: False\n",
"[Step 9/11] Creating infer requests and preparing input tensors\n",
"[ WARNING ] No input files were given for input 'x'!. This input will be filled with random values!\n",
"[ INFO ] Fill input 'x' with random values \n",
"[Step 10/11] Measuring performance (Start inference asynchronously, 1 inference requests, limits: 60000 ms duration)\n",
"[ INFO ] Benchmarking in inference only mode (inputs filling are not included in measurement loop).\n",
"[ INFO ] First inference took 7.94 ms\n",
"[Step 11/11] Dumping statistics report\n",
"[ INFO ] Execution Devices:NPU.3720\n",
"[ INFO ] Count: 17894 iterations\n",
"[ INFO ] Duration: 60004.76 ms\n",
"[ INFO ] Latency:\n",
"[ INFO ] Median: 3.29 ms\n",
"[ INFO ] Average: 3.33 ms\n",
"[ INFO ] Min: 3.21 ms\n",
"[ INFO ] Max: 14.38 ms\n",
"[ INFO ] Throughput: 298.21 FPS\n"
]
}
],
"source": [
"!benchmark_app -m {model_path} -d NPU -hint latency"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "690798af-3c42-468b-ae71-f6c7e15907c7",
"metadata": {},
"source": [
"As you can see from the log entries `[Step 4/11] Reading model files` and `[Step 7/11] Loading the model to the device`, it takes less time to read and compile the model after the initial load."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "511fa8a4-c267-4c7a-9651-42d82cea6386",
"metadata": {},
"source": [
"#### NPU vs CPU with Throughput Hint\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"cell_type": "code",
"execution_count": 26,
"id": "9d570a8b-2cdc-4d04-bf5c-35e9e4f558ee",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[Step 1/11] Parsing and validating input arguments\n",
"[ INFO ] Parsing input parameters\n",
"[Step 2/11] Loading OpenVINO Runtime\n",
"[ INFO ] OpenVINO:\n",
"[ INFO ] Build ................................. 2024.1.0-14992-621b025bef4\n",
"[ INFO ] \n",
"[ INFO ] Device info:\n",
"[ INFO ] CPU\n",
"[ INFO ] Build ................................. 2024.1.0-14992-621b025bef4\n",
"[ INFO ] \n",
"[ INFO ] \n",
"[Step 3/11] Setting device configuration\n",
"[Step 4/11] Reading model files\n",
"[ INFO ] Loading model files\n",
"[ INFO ] Read model took 12.00 ms\n",
"[ INFO ] Original model I/O parameters:\n",
"[ INFO ] Model inputs:\n",
"[ INFO ] x (node: x) : f32 / [...] / [1,3,224,224]\n",
"[ INFO ] Model outputs:\n",
"[ INFO ] x.45 (node: aten::linear/Add) : f32 / [...] / [1,1000]\n",
"[Step 5/11] Resizing model to match image sizes and given batch\n",
"[ INFO ] Model batch size: 1\n",
"[Step 6/11] Configuring input of the model\n",
"[ INFO ] Model inputs:\n",
"[ INFO ] x (node: x) : u8 / [N,C,H,W] / [1,3,224,224]\n",
"[ INFO ] Model outputs:\n",
"[ INFO ] x.45 (node: aten::linear/Add) : f32 / [...] / [1,1000]\n",
"[Step 7/11] Loading the model to the device\n",
"[ INFO ] Compile model took 177.18 ms\n",
"[Step 8/11] Querying optimal runtime parameters\n",
"[ INFO ] Model:\n",
"[ INFO ] NETWORK_NAME: Model2\n",
"[ INFO ] OPTIMAL_NUMBER_OF_INFER_REQUESTS: 4\n",
"[ INFO ] NUM_STREAMS: 4\n",
"[ INFO ] AFFINITY: Affinity.HYBRID_AWARE\n",
"[ INFO ] INFERENCE_NUM_THREADS: 16\n",
"[ INFO ] PERF_COUNT: NO\n",
"[ INFO ] INFERENCE_PRECISION_HINT: <Type: 'float32'>\n",
"[ INFO ] PERFORMANCE_HINT: THROUGHPUT\n",
"[ INFO ] EXECUTION_MODE_HINT: ExecutionMode.PERFORMANCE\n",
"[ INFO ] PERFORMANCE_HINT_NUM_REQUESTS: 0\n",
"[ INFO ] ENABLE_CPU_PINNING: False\n",
"[ INFO ] SCHEDULING_CORE_TYPE: SchedulingCoreType.ANY_CORE\n",
"[ INFO ] MODEL_DISTRIBUTION_POLICY: set()\n",
"[ INFO ] ENABLE_HYPER_THREADING: True\n",
"[ INFO ] EXECUTION_DEVICES: ['CPU']\n",
"[ INFO ] CPU_DENORMALS_OPTIMIZATION: False\n",
"[ INFO ] LOG_LEVEL: Level.NO\n",
"[ INFO ] CPU_SPARSE_WEIGHTS_DECOMPRESSION_RATE: 1.0\n",
"[ INFO ] DYNAMIC_QUANTIZATION_GROUP_SIZE: 0\n",
"[ INFO ] KV_CACHE_PRECISION: <Type: 'float16'>\n",
"[Step 9/11] Creating infer requests and preparing input tensors\n",
"[ WARNING ] No input files were given for input 'x'!. This input will be filled with random values!\n",
"[ INFO ] Fill input 'x' with random values \n",
"[Step 10/11] Measuring performance (Start inference asynchronously, 4 inference requests, limits: 60000 ms duration)\n",
"[ INFO ] Benchmarking in inference only mode (inputs filling are not included in measurement loop).\n",
"[ INFO ] First inference took 31.62 ms\n",
"[Step 11/11] Dumping statistics report\n",
"[ INFO ] Execution Devices:['CPU']\n",
"[ INFO ] Count: 3212 iterations\n",
"[ INFO ] Duration: 60082.26 ms\n",
"[ INFO ] Latency:\n",
"[ INFO ] Median: 65.28 ms\n",
"[ INFO ] Average: 74.60 ms\n",
"[ INFO ] Min: 35.65 ms\n",
"[ INFO ] Max: 157.31 ms\n",
"[ INFO ] Throughput: 53.46 FPS\n"
]
}
],
"source": [
"!benchmark_app -m {model_path} -d CPU -hint throughput"
]
},
{
"cell_type": "code",
"execution_count": 27,
"id": "670aa35c-3a7c-45de-807f-d1c09f2e9f9a",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[Step 1/11] Parsing and validating input arguments\n",
"[ INFO ] Parsing input parameters\n",
"[Step 2/11] Loading OpenVINO Runtime\n",
"[ INFO ] OpenVINO:\n",
"[ INFO ] Build ................................. 2024.1.0-14992-621b025bef4\n",
"[ INFO ] \n",
"[ INFO ] Device info:\n",
"[ INFO ] NPU\n",
"[ INFO ] Build ................................. 2024.1.0-14992-621b025bef4\n",
"[ INFO ] \n",
"[ INFO ] \n",
"[Step 3/11] Setting device configuration\n",
"[Step 4/11] Reading model files\n",
"[ INFO ] Loading model files\n",
"[ INFO ] Read model took 11.50 ms\n",
"[ INFO ] Original model I/O parameters:\n",
"[ INFO ] Model inputs:\n",
"[ INFO ] x (node: x) : f32 / [...] / [1,3,224,224]\n",
"[ INFO ] Model outputs:\n",
"[ INFO ] x.45 (node: aten::linear/Add) : f32 / [...] / [1,1000]\n",
"[Step 5/11] Resizing model to match image sizes and given batch\n",
"[ INFO ] Model batch size: 1\n",
"[Step 6/11] Configuring input of the model\n",
"[ INFO ] Model inputs:\n",
"[ INFO ] x (node: x) : u8 / [N,C,H,W] / [1,3,224,224]\n",
"[ INFO ] Model outputs:\n",
"[ INFO ] x.45 (node: aten::linear/Add) : f32 / [...] / [1,1000]\n",
"[Step 7/11] Loading the model to the device\n",
"[ INFO ] Compile model took 2265.07 ms\n",
"[Step 8/11] Querying optimal runtime parameters\n",
"[ INFO ] Model:\n",
"[ INFO ] DEVICE_ID: \n",
"[ INFO ] ENABLE_CPU_PINNING: False\n",
"[ INFO ] EXECUTION_DEVICES: NPU.3720\n",
"[ INFO ] INFERENCE_PRECISION_HINT: <Type: 'float16'>\n",
"[ INFO ] INTERNAL_SUPPORTED_PROPERTIES: {'CACHING_PROPERTIES': 'RO'}\n",
"[ INFO ] LOADED_FROM_CACHE: False\n",
"[ INFO ] NETWORK_NAME: \n",
"[ INFO ] OPTIMAL_NUMBER_OF_INFER_REQUESTS: 4\n",
"[ INFO ] PERFORMANCE_HINT: PerformanceMode.THROUGHPUT\n",
"[ INFO ] PERFORMANCE_HINT_NUM_REQUESTS: 1\n",
"[ INFO ] PERF_COUNT: False\n",
"[Step 9/11] Creating infer requests and preparing input tensors\n",
"[ WARNING ] No input files were given for input 'x'!. This input will be filled with random values!\n",
"[ INFO ] Fill input 'x' with random values \n",
"[Step 10/11] Measuring performance (Start inference asynchronously, 4 inference requests, limits: 60000 ms duration)\n",
"[ INFO ] Benchmarking in inference only mode (inputs filling are not included in measurement loop).\n",
"[ INFO ] First inference took 7.95 ms\n",
"[Step 11/11] Dumping statistics report\n",
"[ INFO ] Execution Devices:NPU.3720\n",
"[ INFO ] Count: 19080 iterations\n",
"[ INFO ] Duration: 60024.79 ms\n",
"[ INFO ] Latency:\n",
"[ INFO ] Median: 12.51 ms\n",
"[ INFO ] Average: 12.56 ms\n",
"[ INFO ] Min: 6.92 ms\n",
"[ INFO ] Max: 25.80 ms\n",
"[ INFO ] Throughput: 317.87 FPS\n"
]
}
],
"source": [
"!benchmark_app -m {model_path} -d NPU -hint throughput"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "9d00869a-1dad-4d5b-8df8-69cc07a36f98",
"metadata": {},
"source": [
"## Limitations\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "ee1c5d72-5496-4150-84e6-92650ad9d2ab",
"metadata": {},
"source": [
"1. Currently, only the models with static shapes are supported on NPU.\n",
"2. If the path to the model file includes non-Unicode symbols, such as in Chinese, the model cannot be used for inference on NPU. It will return an error."
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "b72be49d-6be4-4594-9f9f-1fe1b1999e6e",
"metadata": {},
"source": [
"## Conclusion\n",
"[back to top ⬆️](#Table-of-contents:)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"id": "bddb4c03-b7b1-48d9-bbdd-ade8a1655e3a",
"metadata": {},
"source": [
"This tutorial demonstrates how easy it is to use NPU in OpenVINO, check its properties, and even tailor the model performance through the different performance hints. \n",
"\n",
"Discover the power of Neural Processing Unit (NPU) with OpenVINO through these interactive Jupyter notebooks: \n",
"##### Introduction\n",
"- [**hello-world**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/hello-world): Start your OpenVINO journey by performing inference on an OpenVINO IR model.\n",
"- [**hello-segmentation**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/hello-segmentation): Dive into inference with a segmentation model and explore image segmentation capabilities.\n",
"\n",
"##### Model Optimization and Conversion\n",
"- [**model-tools**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/model-tools): Discover how to download, convert, and benchmark models from the Open Model Zoo.\n",
"- [**tflite-to-openvino**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/tflite-to-openvino): Learn the process of converting TensorFlow Lite models to OpenVINO IR format.\n",
"- [**yolov7-optimization**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/yolov7-optimization): Optimize the YOLOv7 model for enhanced performance in OpenVINO.\n",
"- [**yolov8-optimization**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/yolov8-optimization): Convert and optimize YOLOv8 models for efficient deployment with OpenVINO.\n",
"\n",
"##### Advanced Computer Vision Techniques\n",
"- [**vision-background-removal**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/vision-background-removal): Implement advanced image segmentation and background manipulation with U^2-Net.\n",
"- [**handwritten-ocr**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/handwritten-ocr): Apply optical character recognition to handwritten Chinese and Japanese text.\n",
"- [**image-inpainting**](https://github.com/openvinotoolkit/openvino_notebooks/blob/latest/notebooks/image-inpainting): Explore the art of image in-painting and restore images with missing parts.\n",
"- [**vehicle-detection-and-recognition**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/vehicle-detection-and-recognition): Use pre-trained models for vehicle detection and recognition in images.\n",
"- [**vision-image-colorization**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/vision-image-colorization): Bring black and white images to life by adding color with neural networks.\n",
"\n",
"##### Real-Time Webcam Applications\n",
"- [**tflite-selfie-segmentation**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/tflite-selfie-segmentation): Apply TensorFlow Lite models for selfie segmentation and background processing.\n",
"- [**object-detection-webcam**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/object-detection-webcam): Experience real-time object detection using your webcam and OpenVINO.\n",
"- [**pose-estimation-webcam**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/pose-estimation-webcam): Perform human pose estimation in real-time with webcam integration.\n",
"- [**action-recognition-webcam**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/action-recognition-webcam): Recognize and classify human actions live with your webcam.\n",
"- [**style-transfer-webcam**](https://github.com/openvinotoolkit/openvino_notebooks/blob/latest/notebooks/style-transfer-webcam): Transform your webcam feed with artistic styles in real-time using pre-trained models.\n",
"- [**3D-pose-estimation-webcam**](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/pose-estimation-webcam): Perform 3D multi-person pose estimation with OpenVINO.\n"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.11"
},
"openvino_notebooks": {
"imageUrl": "",
"tags": {
"categories": [
"API Overview"
],
"libraries": [],
"other": [],
"tasks": [
"Image Classification"
]
}
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|