EdgeTran: Device-Aware Co-Search of Transformers for Efficient Inference on Mobile Edge Platforms

Keyphrases

• Transformer 100%
• Device-independent 100%
• Efficient Inference 100%
• Mobile Edge 100%
• Transformer Model 100%
• Co-search 100%
• Edge Devices 100%
• Edge Platform 100%
• Performance Measures 66%
• Hardware Performance 66%
• Power Draw 66%
• Transformer Architecture 66%
• Latency 33%
• Low Energy 33%
• Energy Consumption 33%
• Accuracy Improvement 33%
• Design Space 33%
• Peak Power 33%
• Search Methods 33%
• Hardware Efficiency 33%
• Challenging Problems 33%
• Block Level 33%
• Energy Peak 33%
• Profilers 33%
• Accuracy Performance 33%
• Consumption Power 33%
• Low Peak Power 33%
• Automated Design 33%
• Low-compute 33%
• Efficient Transformer 33%
• Low Latency 33%
• GPU 33%
• Post-processing Step 33%
• Edge Deployment 33%
• BERT Base 33%
• Large Transformer Models 33%
• Complex Transformer 33%

Computer Science

• Transformer Model 100%
• Performance Measure 50%
• Hardware Performance 50%
• Energy Consumption 25%
• Search Technique 25%
• Postprocessing Step 25%
• Bidirectional Encoder Representations From Transformers 25%
• Graphics Processing Unit 25%