Here we provide a list of example code as a getting started guide on using the MERA™ compiler to deploy and run inference on typical deep neural network models using both PyTorch and TFLite frameworks. Check the corresponding docs for information about the tutorial contents.
- PyTorch Resnet50 on Simulator (
pytorch/resnet50_simulator.py
):
Contains an example on how to deploy and run a traced resnet50
model in x86 host simulation.
Can be executed with the following command:
cd tutorials/pytorch
python3 resnet50_simulator.py
- PyTorch Resnet50 on IP (
pytorch/resnet50_ip.py
):
Contains an example on how to deploy and run a traced resnet50
model in FPGA environment.
Needs to have FPGA runtime setup before running.
Can be executed with the following command:
cd tutorials/pytorch
# Needs to enable RUN_IP env in order to actually run the tutorial in HW
RUN_IP=1 python3 resnet50_ip.py
- TFLite EfficientNet on Simulator (
tflite/efficientnet_simulator.py
):
Contains an example on how to deploy and run a quantized efficientnet-lite1
and efficientnet-lite4
model in x86 host simulation and run an example object classification.
Can be executed with the following command:
cd tutorials/tflite
python3 efficientnet_simulator.py
- TFLite EfficientNet on IP (
tflite/efficientnet_ip.py
):
Contains an example on how to deploy and run a quantized efficientnet-lite1
and efficientnet-lite4
model in
FPGA environment and run an example object classification. Needs to have FPGA runtime setup before running.
Can be executed with the following command:
cd tutorials/tflite
# Needs to enable RUN_IP env in order to actually run the tutorial in HW
RUN_IP=1 python3 efficientnet_ip.py
- Fused PyTorch Resnet18 + MobilenetV2 on Simulator (
multi_models/fused_resnet_mobilenet_simulator.py
):
Contains an example on how to fuse two quantized PyTorch models (i.e., resnet18
and mobilenet_v2
) and then deploy the fused model in x86 host simulation.
Can be executed with the following command:
cd tutorials/multi_models
python3 fused_resnet_mobilenet_simulator.py