| trtexec=/usr/src/tensorrt/bin/trtexec | |
| export CUDA_VISIBLE_DEVICES="0" | |
| MIN_BATCH=1 | |
| OPT_BATCH=4 | |
| MAX_BATCH=16 | |
| onnx_path=./model_repo_sense_voice_small/encoder/1/model.onnx | |
| $trtexec --onnx=$onnx_path --saveEngine=model.plan \ | |
| --minShapes=speech:${MIN_BATCH}x8x560,speech_lengths:${MIN_BATCH},language:${MIN_BATCH},textnorm:${MIN_BATCH} \ | |
| --optShapes=speech:${OPT_BATCH}x128x560,speech_lengths:${OPT_BATCH},language:${OPT_BATCH},textnorm:${OPT_BATCH} \ | |
| --maxShapes=speech:${MAX_BATCH}x512x560,speech_lengths:${MAX_BATCH},language:${MAX_BATCH},textnorm:${MAX_BATCH} | |
| $trtexec --loadEngine=model.plan --noDataTransfers --verbose \ | |
| --shapes=speech:${OPT_BATCH}x65x560,speech_lengths:${OPT_BATCH},language:${OPT_BATCH},textnorm:${OPT_BATCH} |