Welcome to the XuanTie AI Benchmark Suite! This repository contains a comprehensive collection of benchmark tests specifically designed for RISC-V CPUs with RVV/RVM. Our goal is to provide developers, researchers, and enthusiasts with a suite of tools to evaluate the performance and efficiency of AI algorithms running on RISC-V architecture.
Model | Original Model | License | quantization | c907 | c908 |
---|---|---|---|---|---|
bert | url | Apache-2.0 license | float32 | url | url |
float16 | url | url | |||
int8_asym_w_sym | url | url | |||
float16_w_int8 | url | ||||
efficientnet | url | Apache-2.0 license | float32 | url | url |
float16 | url | url | |||
int8_asym_w_sym | url | url | |||
float16_w_int8 | url | ||||
mobilenetv2 | url | Apache-2.0 license | float32 | url | url |
float16 | url | url | |||
int8_asym_w_sym | url | url | |||
float16_w_int8 | url | ||||
mobileVit | url | apple | float32 | url | url |
float16 | url | url | |||
int8_asym_w_sym | url | url | |||
float16_w_int8 | url | ||||
resnet50 | url | Apache-2.0 license | float32 | url | url |
float16 | url | url | |||
int8_asym_w_sym | url | url | |||
float16_w_int8 | url | ||||
retinaface | url | MIT | float32 | url | url |
float16 | url | url | |||
int8_asym_w_sym | url | url | |||
float16_w_int8 | url | ||||
shufflenetv2 | url | Apache-2.0 license | float32 | url | url |
float16 | url | url | |||
int8_asym_w_sym | |||||
float16_w_int8 | |||||
swin | url | MIT | float32 | url | url |
float16 | url | url | |||
int8_asym_w_sym | url | url | |||
float16_w_int8 | url | ||||
yolov5n | url | AGPL-3.0 | float32 | ✓ | ✓ |
float16 | ✓ | ✓ | |||
int8_asym_w_sym | ✓ | ✓ | |||
float16_w_int8 | ✓ |
- riscv gcc: https://www.xrvm.cn/community/download?id=4333581795569242112
- qemu(not necessary): https://www.xrvm.cn/community/download?id=4300747789186437120
# 1. download codes
git clone repo_url
# 2. prepare submodules
cd REPO_DIR
git submodule init
git submodule update
Take c908/float16/onnx_mobilenetv2-12
as an example
cd c908/float16/onnx_mobilenetv2-12
make -f makefile.c908 hhb_runtime
# download binary file for model(click the url in the above table or wget)
wget http://yocbook.oss-cn-hangzhou.aliyuncs.com/linux_image/XuanTie_V1.0.5/xtai-benchmark-bm/908_float16_onnx_mobilenetv2-12_hhb.bm.tar.gz
tar -zxvf 908_float16_onnx_mobilenetv2-12_hhb.bm.tar.gz
mv 908_float16_onnx_mobilenetv2-12_hhb.bm hhb.bm
# run with qemu
make -f makefile.c908 run_sim