First release version, support 3 perf mode currently:
- byte_micro_perf
- op level perf, add/gemm/allreduce etc.
- python3 ./launch.py --task add,gemm,allreduce --hardware_type GPU
- byte_infer_perf/general_perf
- model level perf, resnet50/yolov5 etc.
- python3 ./launch.py --task resnet50-torch-fp32 --hardware_type CPU
- byte_infer_perf/llm_perf
- llm model level perf, llama3-70b etc.
- python3 ./launch.py --task llama3-torch-bf16-70b --hardware_type GPU