YAML Metadata Warning: empty or missing yaml metadata in repo card (https://huggingface.co/docs/hub/model-cards#model-card-metadata)

MobileNet v2 1.0 224 Optimized for ARM Ethos-U55

This repository contains MobileNet v2 1.0 224 model optimized for the ARM Ethos-U55 NPU using the Vela compiler.

Model Overview

  • Base Model: MobileNet v2 1.0 224 INT8 quantized
  • Source: ARM ML-zoo repository
  • Target Hardware: ARM Ethos-U55 NPU
  • Optimization: Compiled with Vela compiler for optimal performance

Files Description

Model Files

  • mobilenet_v2_1.0_224_INT8_vela.tflite - Vela-optimized model for Ethos-U55
  • mobilenet_v2_1.0_224_INT8.tflite - Original INT8 quantized model
  • labelmappings.txt - ImageNet class labels (1001 classes)

Configuration Files

  • u55_eval_with_TA_config_400_and_200_MHz.ini - Vela configuration for U55 evaluation

Performance Reports

  • mobilenet_v2_1.0_224_INT8_summary_Ethos_U55_400MHz_SRAM_3.2_GBs_Flash_0.05_GBs.csv - Performance summary
  • mobilenet_v2_1.0_224_INT8_per-layer.csv - Per-layer performance analysis

Performance Summary

System Configuration: Ethos_U55_400MHz_SRAM_3.2_GBs_Flash_0.05_GBs

  • Total SRAM used: 353.50 KiB
  • Total On-chip Flash used: 3614.39 KiB
  • NPU cycles: 6,019,265 cycles/batch
  • Network Tops/s: 0.04 Tops/s
  • CPU operators: 0 (0.0%)
  • NPU operators: 95 (100.0%)

Model Details

  • Input Shape: (224, 224, 3)
  • Quantization: INT8 weights and activations
  • MAC Operations: 304,452,946 MACs/batch
  • All operators: Compatible with Ethos-U55

Usage

The Vela-optimized model (mobilenet_v2_1.0_224_INT8_vela.tflite) is ready for deployment on ARM Ethos-U55 systems.

Compilation Command

The model was compiled using the following Vela command:

vela --accelerator-config=ethos-u55-128 \
     --optimise Size \
     --config u55_eval_with_TA_config_400_and_200_MHz.ini \
     --memory-mode Sram_Only \
     --system-config Ethos_U55_400MHz_SRAM_3.2_GBs_Flash_0.05_GBs \
     mobilenet_v2_1.0_224_INT8.tflite \
     --verbose-cycle-estimate \
     --verbose-performance \
     --output-dir vela_output

References

Downloads last month
2
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support