Open Pre-Trained Models

Usage - Compile and run

Install tpu-perf to build and run model cases.

# Time only cases
python3 -m tpu_perf.build --list default_cases.txt --time
python3 -m tpu_perf.run --list default_cases.txt

# Precision benchmark
python3 -m tpu_perf.build --list default_cases.txt
python3 -m tpu_perf.precision_benchmark --list default_cases.txt

Usage - Git LFS

On default, cloning this repository will not download any models. Install Git LFS with pip install git-lfs.

To download a specific model: git lfs pull --include="path/to/model" --exclude=""

To download all models: git lfs pull --include="*" --exclude=""

Usage - Model visualization

You can see visualizations of each model's network architecture by using Netron.

How to contribute

Please lint in your local repo before PR.

# Install tools
sudo npm install -g markdownlint-cli
pip3 install yamllint

yamllint -c ./.yaml-lint.yml .
markdownlint '**/*.md'
python3 .github/workflows/check.py

Model Navigation

Vision

Model	Path	NNTC	MLIR
alexnet	vision/classification/AlexNet-Caffe	✅
arcface	vision/recognition/arcface		✅
big_transfer	vision/classification/big_transfer_mlir		✅
BiSeNet	vision/segmentation/BiSeNet		✅
c3d	vision/video-recognition/C3D	✅	✅
CenterNet	vision/detection/CenterNet-Torch	✅
clip	vision/classification/clip-mlir		✅
CRNN	vision/OCR/CRNN	✅
cyclegan_horse2zebra	vision/GAN/cyclegan	✅
dbnet	vision/detection/dbnet	✅	✅
DBNet_totaltext_res18_dcn	vision/OCR/DBNet	✅
DBNet_totaltext_res50_dcn	vision/OCR/DBNet	✅
deeplabv3p	vision/segmentation/deeplabv3p	✅	✅
densenet	vision/classification/DenseNet-Caffe	✅
dpn68	vision/classification/dpn68	✅	✅
ECANet	vision/classification/ECANet-Torch	✅
efficientnet-B0	vision/classification/EfficientNet-B0	✅
efficientnet-B1	vision/classification/EfficientNet-B1	✅
efficientnet-B4	vision/classification/EfficientNet-B4	✅
efficientnet-B5	vision/classification/EfficientNet-B5	✅
efficientnet-B7	vision/classification/EfficientNet-B7	✅
efficientnet-lite4	vision/classification/efficientnet-lite4		✅
efficientnetv2	vision/classification/efficientnetv2	✅	✅
ERFNet	vision/segmentation/ERFNet-Caffe	✅
face_alignment	vision/recognition/face_alignment		✅
facenet	vision/recognition/facenet	✅
GOTURN	vision/tracking/GOTURN-Caffe	✅
HRNet	vision/classification/HRNet-Torch	✅
huggingface-resnet50	vision/classification/resnet50-v1.5	✅
inception_resnet_v2	vision/classification/inception_resnet_v2	✅
inception_v1	vision/classification/inception_v1	✅
inception_v3	vision/classification/inception_v3	✅
inception_v4	vision/classification/inception_v4	✅
inception_v4-tflite	vision/classification/inception_v4		✅
inceptionBN21k	vision/classification/InceptionBN-21k-for-Caffe	✅
lenet	vision/classification/LeNet	✅
lprnet	vision/detection/lprnet		✅
market_bot_R50	vision/reid/market_bot_R50	✅	✅
MDNet	vision/tracking/MDNet-Torch	✅	✅
mm_resnet50	vision/classification/mm_resnet50	✅	✅
mobilenet-v2	vision/classification/mobilenet-v2		✅
mobilenet_v1	vision/classification/MobileNet-Caffe-v1	✅
mobilenetv2	vision/classification/MobileNet-Caffe-v2	✅	✅
mobilenetv3	vision/classification/MobileNet-Caffe-v3	✅
mobileseg	vision/segmentation/mobileseg	✅
mobileseg-mlir	vision/segmentation/mobileseg		✅
mtcnn_onet	vision/detection/mtcnn	✅	✅
mtcnn_pnet	vision/detection/mtcnn	✅	✅
mtcnn_rnet	vision/detection/mtcnn	✅	✅
openpose	vision/pose-estimation/openpose	✅	✅
pointpillars	vision/detection/pointpillars		✅
PP-OCRv3_det	vision/OCR/OCRv3_paddle		✅
PP-OCRv3_rec	vision/OCR/OCRv3_paddle		✅
PP-OCRv3cls	vision/OCR/PP-OCRv3cls	✅	✅
PP-OCRv3det	vision/OCR/PP-OCRv3det	✅	✅
PP-OCRv3rec	vision/OCR/PP-OCRv3rec		✅
pp_humansegv1_lite_argmax	vision/segmentation/pp_humansegv1_lite	✅
pp_humansegv1_lite_softmax	vision/segmentation/pp_humansegv1_lite	✅
pp_humansegv2_mobile	vision/segmentation/pp_humansegv2_mobile		✅
pp_liteseg	vision/segmentation/pp_liteseg	✅	✅
pp_picodet_s	vision/detection/pp-picodet		✅
ppocr_mobile_v2.0_cls	vision/OCR/OCRv3_paddle		✅
ppyoloe_crn_s_300e_coco	vision/detection/ppyoloe		✅
ppyoloe_crn_x_300e_coco	vision/detection/ppyoloe	✅
ppyoloe_plus_crn_x_80e_coco	vision/detection/ppyoloe	✅
ppyolov2_r101vd_dcn_365e_coco	vision/detection/ppyolo	✅
py-R-FCN	vision/detection/py-R-FCN	✅
res2net101_26w_4s	vision/classification/res2net101_26w_4s	✅	✅
res2net50_26w_4s	vision/classification/res2net50_26w_4s	✅	✅
resnet101-v1-7	vision/classification/resnet101-v1-7		✅
resnet152-v1-7	vision/classification/resnet152-v1-7		✅
resnet18-v1-7	vision/classification/resnet18-v1-7	✅	✅
resnet18-v2	vision/classification/resnet18-v2		✅
resnet34	vision/classification/ResNet34	✅
resnet34-v1-7	vision/classification/resnet34-v1-7	✅	✅
resnet50-caffe	vision/classification/ResNet50-Caffe	✅
resnet50-v1-7	vision/classification/resnet50-v1-7	✅	✅
resnet50-v2	vision/classification/resnet50-v2		✅
ResNet50_vd_infer	vision/classification/ResNet50_vd_paddle		✅
resnext	vision/classification/ResNeXt	✅
resneXt50	vision/classification/ResNeXt50	✅
retinaface	vision/detection/retinaface	✅
scrfd	vision/detection/scrfd		✅
SegFormer	vision/segmentation/SegFormer		✅
shufflenet_v2	vision/classification/shufflenet_v2		✅
shufflenetv2	vision/classification/shufflenet_v2_torch	✅
SiamMask	vision/tracking/SiamMask-Torch	✅	✅
squeezenet	vision/classification/SqueezeNet	✅
squeezenet1.0	vision/classification/squeezenet1.0		✅
SRCNN	vision/super-resolution/SRCNN	✅
ssd-mobilenet-tflite	vision/detection/ssd-mobilenet		✅
StableDiffusion_decode	vision/generative/stable_diffusion		✅
StableDiffusion_text_encoder	vision/generative/stable_diffusion		✅
StableDiffusion_unet	vision/generative/stable_diffusion		✅
swin_t	vision/classification/SwinTransformer	✅	✅
tpu-mlir_S-DCNet_SHA	vision/visual-counting/S-DCNet		✅
tpu-mlir_S-DCNet_SHB	vision/visual-counting/S-DCNet		✅
tpu-mlir_yoloface	vision/detection/yoloface		✅
tsm	vision/recognition/tsm		✅
TSN	vision/video-recognition/TSN		✅
ultraface	vision/detection/ultraface		✅
ultralytics_yolov3	vision/detection/ultralytics-yolov3	✅
unet_plusplus	vision/segmentation/unet_plusplus	✅	✅
VDSR	vision/super-resolution/VDSR	✅
vgg11	vision/classification/vgg11-torch	✅
vgg16	vision/classification/vgg16		✅
vgg19	vision/classification/vgg19	✅
vggssd_300	vision/detection/vggssd_300	✅
vision_OCR_CRNN_tpu-mlir	vision/OCR/CRNN		✅
vit-base-patch16-384	vision/classification/vit_B_16	✅
vit-large-patch16-384	vision/classification/vit_L_16	✅
WRN-50-2	vision/classification/WRN-50-2	✅	✅
wrn50	vision/classification/wrn50	✅	✅
xception	vision/classification/xception	✅
Yet-Another-EfficientDet-Pytorch	vision/detection/Yet-Another-EfficientDet-Pytorch	✅
yolov3	vision/detection/yolov3-torch	✅
yolov3_320	vision/detection/yolov3_320	✅
yolov3_608	vision/detection/yolov3_608	✅
yolov3_mobilenet_v3_270e_coco	vision/detection/ppyolov3		✅
yolov3_spp	vision/detection/yolov3_spp	✅
yolov3_tiny	vision/detection/yolov3_tiny	✅
yolov5l	vision/detection/yolov5		✅
yolov5l-5	vision/detection/yolov5	✅
yolov5m	vision/detection/yolov5		✅
yolov5m-5	vision/detection/yolov5	✅
yolov5m-6	vision/detection/yolov5	✅
yolov5s	vision/detection/yolov5		✅
yolov5s-4	vision/detection/yolov5	✅
yolov5s-5	vision/detection/yolov5	✅
yolov5s-tflite	vision/detection/yolov5		✅
yolov5x	vision/detection/yolov5		✅
yolov5x-5	vision/detection/yolov5	✅
yolox_s_300e_coco	vision/detection/ppyolox		✅

Language

Model	Path	NNTC	MLIR
bert	language/nlp/bert	✅
bert_base_transformers-2.11.0	language/nlp/Huggingface_bert_squadv1	✅
bert_base_transformers-4.23.0	language/nlp/Huggingface_bert_squadv1	✅
bert_large_transformers-2.11.0	language/nlp/Huggingface_bert_squadv1	✅
bert_large_transformers-4.23.0	language/nlp/Huggingface_bert_squadv1	✅
bert_paddle	language/nlp/bert_paddle	✅
ConformerEncoder	language/asr/conformer	✅
gpt2	language/generative/gpt2	✅
GRU	language/nlp/GRU	✅	✅
mobilebert-tflite	language/nlp/mobilebert_tflite		✅
ScoringIds	language/asr/conformer	✅
TransformerDecoder	language/asr/conformer	✅
TransformerLM	language/asr/conformer	✅
WenetDecoder	language/asr/wenet	✅
WenetEncoder	language/asr/wenet	✅