Commit
3 years ago
ROCm MHA (#15279) Add MultiHeadAttention for ROCm EP. **Before:** ``` 'engine': 'onnxruntime' 'version': '1.15.0' 'height': 512 'width': 512 'steps': 50 'batch_size': 1 'batch_count': 5 'num_prompts': 1 'average_latency': 3.878769588470459 'median_latency': 3.8792178630828857 'first_run_memory_MB': -1 'second_run_memory_MB': -1 'model_name': 'runwayml/stable-diffusion-v1-5' 'directory': './sd-v1-5-onnx-fp16-nomha' 'provider': 'ROCMExecutionProvider' 'disable_safety_checker': True ``` **After:** ``` 'engine': 'onnxruntime' 'version': '1.15.0' 'height': 512 'width': 512 'steps': 50 'batch_size': 1 'batch_count': 5 'num_prompts': 1 'average_latency': 2.364924430847168 'median_latency': 2.3650705814361572 'first_run_memory_MB': -1 'second_run_memory_MB': -1 'model_name': 'runwayml/stable-diffusion-v1-5' 'directory': './sd-v1-5-onnx-fp16' 'provider': 'ROCMExecutionProvider' 'disable_safety_checker': True ```
Author
Parents
Loading