Skip to main content

ResNet50

Guide for running the ResNet50 image classification model on BOS Eagle-N hardware. Covers dataset preparation, model execution with various batch sizes, performance optimization options, and profiling with Tracy and ttnn-visualizer tools.

(Optional) Download ImageNet-1K(validation set)​

unzip imagenet-val.zip -d imagenet-val

Set environment variables​

# at $TT_METAL_HOME
source env_set.sh

Run ResNet50 (Image)​

python models/bos_model/resnet50/run_resnet50.py [--device_id $DEVICE_ID] [--batch $BATCH_SIZE] [-n $ITER] [--trace] [--cq2] [--data_dir /path/to/imagenet-1k] [--seed $SEED_VALUE] [--no_shuffle] [--demo] [--fullscreen] [--delay $DELAY_TIME] [--benchmark]
  • --device_id: If you have multi device, you can choose device for run model($DEVICE_ID = 0 is default)
  • -b or --batch: You can choose batch size(B = [1, 2, 4(default)])
  • -n: You can run not only 1 time but also $ITER times(default is 1)
  • --trace: If you use this option, you can use trace(makes model execution much more faster)
  • --cq2: If you use this option with --trace, you can use 2 command queue(makes model execution more faster)
  • --data_dir: You can choose dataset(XXX = [models/bos_model/demo/dataset/sample/(default), /path/to/imagenet-1k])
  • --seed: If you want to shuffle dataset using another random seed, you can use another integer value($SEED_VALUE = 0 is default)
  • --no_shuffle: If you don't want to shuffle dataset, use this option
  • --demo: If you use this option, you can see demo window (default: only result show on console)
  • --fullscreen: If you use this option, you can see full size demo window
  • -delay: You can run with delay($DELAY_TIME is floating number, default is 0.0)
  • --benchmark: If you use this option, you can see model's e2e performance which called FPS (you can see FPS also using --demo)

Example: Run Demo (with Batch 4, ImageNet-1K, Trace, 2CQ, Visualization+Full Screen, 1 Time)​

python models/bos_model/resnet50/run_resnet50.py --data_dir models/bos_model/demo/dataset/imagenet-val/ --trace --cq2 --demo --fullscreen

Run with Tracy Profiler​

python -m tracy -r -p -v -m pytest models/bos_model/resnet50/tests/test_ttnn_functional_resnet50.py

Run for ttnn-visualizer Profiler​

  • First, export ENV using script file
    • $EXPERIMENT_NAME: input anythings (for example, resnet)
source models/bos_model/export_l1_vis.sh $EXPERIMENT_NAME
  • Second, run model
    • If the model has finished running successfully, the result report will be generated in the following path (generated/ttnn/reports/$EXPERIMENT_NAME_MMDD_hhmm/)
pytest models/bos_model/resnet50/tests/test_ttnn_functional_resnet50.py -k batch_4
  • Third, run ttnn-visualizer
    • $REPORT_PATH: It is the path mentioned in the previous step
    • visit http://localhost:8000/ using your web-browser
ttnn-visualizer --profiler-path $REPORT_PATH
  • If the experiment has finished, please run the following command to clear the environment variables
source models/bos_model/unset_l1_vis.sh