官方模型库
领域套件与扩展包
计算机视觉
图像分类(骨干类)
以下数据基于Atlas训练系列产品和ImageNet-1K数据集获得。
model |
acc@1 |
mindcv recipe |
vanilla mindspore |
---|---|---|---|
vgg11 |
71.86 |
||
vgg13 |
72.87 |
||
vgg16 |
74.61 |
||
vgg19 |
75.21 |
||
resnet18 |
70.21 |
||
resnet34 |
74.15 |
||
resnet50 |
76.69 |
||
resnet101 |
78.24 |
||
resnet152 |
78.72 |
||
resnetv2_50 |
76.90 |
||
resnetv2_101 |
78.48 |
||
dpn92 |
79.46 |
||
dpn98 |
79.94 |
||
dpn107 |
80.05 |
||
dpn131 |
80.07 |
||
densenet121 |
75.64 |
||
densenet161 |
79.09 |
||
densenet169 |
77.26 |
||
densenet201 |
78.14 |
||
seresnet18 |
71.81 |
||
seresnet34 |
75.36 |
||
seresnet50 |
78.31 |
||
seresnext26 |
77.18 |
||
seresnext50 |
78.71 |
||
skresnet18 |
73.09 |
||
skresnet34 |
76.71 |
||
skresnet50_32x4d |
79.08 |
||
resnext50_32x4d |
78.53 |
||
resnext101_32x4d |
79.83 |
||
resnext101_64x4d |
80.30 |
||
resnext152_64x4d |
80.52 |
||
rexnet_x09 |
77.07 |
||
rexnet_x10 |
77.38 |
||
rexnet_x13 |
79.06 |
||
rexnet_x15 |
79.94 |
||
rexnet_x20 |
80.64 |
||
resnest50 |
80.81 |
||
resnest101 |
82.50 |
||
res2net50 |
79.35 |
||
res2net101 |
79.56 |
||
res2net50_v1b |
80.32 |
||
res2net101_v1b |
95.41 |
||
googlenet |
72.68 |
||
inceptionv3 |
79.11 |
||
inceptionv4 |
80.88 |
||
mobilenet_v1_025 |
53.87 |
||
mobilenet_v1_050 |
65.94 |
||
mobilenet_v1_075 |
70.44 |
||
mobilenet_v1_100 |
72.95 |
||
mobilenet_v2_075 |
69.98 |
||
mobilenet_v2_100 |
72.27 |
||
mobilenet_v2_140 |
75.56 |
||
mobilenet_v3_small |
68.10 |
||
mobilenet_v3_large |
75.23 |
||
shufflenet_v1_g3_x0_5 |
57.05 |
||
shufflenet_v1_g3_x1_5 |
67.77 |
||
shufflenet_v2_x0_5 |
57.05 |
||
shufflenet_v2_x1_0 |
67.77 |
||
shufflenet_v2_x1_5 |
57.05 |
||
shufflenet_v2_x2_0 |
67.77 |
||
xception |
79.01 |
||
ghostnet_50 |
66.03 |
||
ghostnet_100 |
73.78 |
||
ghostnet_130 |
75.50 |
||
nasnet_a_4x1056 |
73.65 |
||
mnasnet_0.5 |
68.07 |
||
mnasnet_0.75 |
71.81 |
||
mnasnet_1.0 |
74.28 |
||
mnasnet_1.4 |
76.01 |
||
efficientnet_b0 |
76.89 |
||
efficientnet_b1 |
78.95 |
||
efficientnet_b2 |
79.80 |
||
efficientnet_b3 |
80.50 |
||
efficientnet_v2 |
83.77 |
||
regnet_x_200mf |
68.74 |
||
regnet_x_400mf |
73.16 |
||
regnet_x_600mf |
73.34 |
||
regnet_x_800mf |
76.04 |
||
regnet_y_200mf |
70.30 |
||
regnet_y_400mf |
73.91 |
||
regnet_y_600mf |
75.69 |
||
regnet_y_800mf |
76.52 |
||
mixnet_s |
75.52 |
||
mixnet_m |
76.64 |
||
mixnet_l |
78.73 |
||
hrnet_w32 |
80.64 |
||
hrnet_w48 |
81.19 |
||
bit_resnet50 |
76.81 |
||
bit_resnet50x3 |
80.63 |
||
bit_resnet101 |
77.93 |
||
repvgg_a0 |
72.19 |
||
repvgg_a1 |
74.19 |
||
repvgg_a2 |
76.63 |
||
repvgg_b0 |
74.99 |
||
repvgg_b1 |
78.81 |
||
repvgg_b2 |
79.29 |
||
repvgg_b3 |
80.46 |
||
repvgg_b1g2 |
78.03 |
||
repvgg_b1g4 |
77.64 |
||
repvgg_b2g4 |
78.80 |
||
repmlp_t224 |
76.71 |
||
convnext_tiny |
81.91 |
||
convnext_small |
83.40 |
||
convnext_base |
83.32 |
||
vit_b_32_224 |
75.86 |
||
vit_l_16_224 |
76.34 |
||
vit_l_32_224 |
73.71 |
||
swintransformer_tiny |
80.82 |
||
pvt_tiny |
74.81 |
||
pvt_small |
79.66 |
||
pvt_medium |
81.82 |
||
pvt_large |
81.75 |
||
pvt_v2_b0 |
71.50 |
||
pvt_v2_b1 |
78.91 |
||
pvt_v2_b2 |
81.99 |
||
pvt_v2_b3 |
82.84 |
||
pvt_v2_b4 |
83.14 |
||
pit_ti |
72.96 |
||
pit_xs |
78.41 |
||
pit_s |
80.56 |
||
pit_b |
81.87 |
||
coat_lite_tiny |
77.35 |
||
coat_lite_mini |
78.51 |
||
coat_tiny |
79.67 |
||
convit_tiny |
73.66 |
||
convit_tiny_plus |
77.00 |
||
convit_small |
81.63 |
||
convit_small_plus |
81.80 |
||
convit_base |
82.10 |
||
convit_base_plus |
81.96 |
||
crossvit_9 |
73.56 |
||
crossvit_15 |
81.08 |
||
crossvit_18 |
81.93 |
||
mobilevit_xx_small |
68.90 |
||
mobilevit_x_small |
74.98 |
||
mobilevit_small |
78.48 |
||
visformer_tiny |
78.28 |
||
visformer_tiny_v2 |
78.82 |
||
visformer_small |
81.76 |
||
visformer_small_v2 |
82.17 |
||
edgenext_xx_small |
71.02 |
||
edgenext_x_small |
75.14 |
||
edgenext_small |
79.15 |
||
edgenext_base |
82.24 |
||
poolformer_s12 |
77.33 |
||
xcit_tiny_12_p16 |
77.67 |
目标检测
以下数据基于Atlas训练系列产品和COCO2017数据集获得。
yolo
model |
map |
mindyolo recipe |
vanilla mindspore |
---|---|---|---|
yolov8_n |
37.2 |
||
yolov8_s |
44.6 |
||
yolov8_m |
50.5 |
||
yolov8_l |
52.8 |
||
yolov8_x |
53.7 |
||
yolov7_t |
37.5 |
||
yolov7_l |
50.8 |
||
yolov7_x |
52.4 |
||
yolov5_n |
27.3 |
||
yolov5_s |
37.6 |
||
yolov5_m |
44.9 |
||
yolov5_l |
48.5 |
||
yolov5_x |
50.5 |
||
yolov4_csp |
45.4 |
||
yolov4_csp(silu) |
45.8 |
||
yolov3_darknet53 |
45.5 |
||
yolox_n |
24.1 |
||
yolox_t |
33.3 |
||
yolox_s |
40.7 |
||
yolox_m |
46.7 |
||
yolox_l |
49.2 |
||
yolox_x |
51.6 |
||
yolox_darknet53 |
47.7 |
经典
model |
map |
mind_series recipe |
vanilla mindspore |
---|---|---|---|
ssd_vgg16 |
23.2 |
||
ssd_mobilenetv1 |
22.0 |
||
ssd_mobilenetv2 |
29.1 |
||
ssd_resnet50 |
34.3 |
||
fasterrcnn |
58 |
||
maskrcnn_mobilenetv1 |
coming soon |
||
maskrcnn_resnet50 |
coming soon |
语义分割
model |
mind_series recipe |
vanilla mindspore |
---|---|---|
ocrnet |
||
deeplab v3 |
||
deeplab v3 plus |
||
unet |
OCR
文本检测
model |
dataset |
fscore |
mindocr recipe |
vanilla mindspore |
---|---|---|---|---|
dbnet_mobilenetv3 |
icdar2015 |
77.23 |
||
dbnet_resnet18 |
icdar2015 |
81.73 |
||
dbnet_resnet50 |
icdar2015 |
85.05 |
||
dbnet++_resnet50 |
icdar2015 |
86.74 |
||
psenet_resnet152 |
icdar2015 |
82.06 |
||
east_resnet50 |
icdar2015 |
84.87 |
||
fcenet_resnet50 |
icdar2015 |
84.12 |
文本识别
model |
dataset |
acc |
mindocr recipe |
vanilla mindspore |
---|---|---|---|---|
svtr_tiny |
IC03,13,15,IIIT,etc |
89.02 |
||
crnn_vgg7 |
IC03,13,15,IIIT,etc |
82.03 |
||
crnn_resnet34_vd |
IC03,13,15,IIIT,etc |
84.45 |
||
rare_resnet34_vd |
IC03,13,15,IIIT,etc |
85.19 |
文本方向分类
model |
dataset |
acc |
mindocr recipe |
---|---|---|---|
mobilenetv3 |
RCTW17,MTWI,LSVT |
94.59 |
人脸
model |
dataset |
acc |
mindface recipe |
vanilla mindspore |
---|---|---|---|---|
arcface_mobilefacenet-0.45g |
MS1MV2 |
98.70 |
||
arcface_r50 |
MS1MV2 |
99.76 |
||
arcface_r100 |
MS1MV2 |
99.38 |
||
arcface_vit_t |
MS1MV2 |
99.71 |
||
arcface_vit_s |
MS1MV2 |
99.76 |
||
arcface_vit_b |
MS1MV2 |
99.81 |
||
arcface_vit_l |
MS1MV2 |
99.75 |
||
retinaface_mobilenet_0.25 |
WiderFace |
90.77/88.2/74.76 |
||
retinaface_r50 |
WiderFace |
95.07/93.61/84.84 |
语言模型
model |
mindformer recipe |
vanilla mindspore |
---|---|---|
bert_base |
||
t5_small |
||
gpt2_small |
||
gpt2_13b |
||
gpt2_52b |
||
pangu_alpha |
||
glm_6b |
||
glm_6b_lora |
||
llama_7b |
||
llama_13b |
||
llama_65b |
||
llama_7b_lora |
||
bloom_560m |
||
bloom_7.1b |
||
bloom_65b |
||
bloom_176b |
强化学习
Algorithm |
Discrete Action Space |
Continuous Action Space |
CPU |
GPU |
Ascend |
Environment |
Reward |
---|---|---|---|---|---|---|---|
✅ |
/ |
✅ |
✅ |
✅ |
195 |
||
/ |
✅ |
✅ |
✅ |
✅ |
4800 |
||
✅ |
/ |
✅ |
✅ |
✅ |
195 |
||
✅ |
/ |
✅ |
✅ |
✅ |
195 |
||
/ |
✅ |
✅ |
✅ |
✅ |
4800 |
||
✅ |
/ |
✅ |
✅ |
✅ |
90%/-145 |
||
/ |
✅ |
✅ |
✅ |
✅ |
4800 |
||
/ |
✅ |
✅ |
✅ |
✅ |
4800 |
||
✅ |
/ |
✅ |
✅ |
✅ |
195 |
||
✅ |
/ |
/ |
✅ |
✅ |
195 |
||
/ |
✅ |
✅ |
✅ |
✅ |
3500 |
||
✅ |
/ |
✅ |
✅ |
✅ |
-145 |
||
/ |
✅ |
✅ |
✅ |
✅ |
4800 |
||
/ |
✅ |
✅ |
✅ |
✅ |
5000 |
||
/ |
✅ |
/ |
✅ |
✅ |
900 |
||
/ |
✅ |
✅ |
✅ |
✅ |
3000 |
||
✅ |
/ |
✅ |
✅ |
✅ |
-140 |
||
✅ |
/ |
✅ |
✅ |
✅ |
195 |
||
✅ |
/ |
✅ |
✅ |
✅ |
195 |
||
✅ |
/ |
✅ |
✅ |
✅ |
195 |
科学计算套件
领域 |
网络 |
MindSpore实现 |
Ascend |
GPU |
---|---|---|---|---|
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
|||
通用物理 |
✅ |
|||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
|||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
|||
通用物理 |
✅ |
|||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
✅ |
||
通用物理 |
✅ |
✅ |
||
哈密顿系统 |
✅ |
✅ |
||
弹性动力学 |
✅ |
✅ |
||
热力学 |
✅ |
✅ |
||
气象学 |
✅ |
✅ |
||
地质学 |
✅ |
✅ |
||
地质学 |
✅ |
✅ |
||
海洋物理 |
✅ |
|||
海洋物理 |
✅ |
✅ |
||
电磁学 |
✅ |
✅ |
||
电磁学 |
✅ |
✅ |
||
电磁学 |
✅ |
✅ |
||
电磁学 |
✅ |
✅ |
||
电磁学 |
✅ |
|||
电磁学 |
✅ |
✅ |
||
电磁学 |
✅ |
✅ |
||
电磁学 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算生物 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
||
计算流体 |
✅ |
✅ |
大模型套件
Transformers
model |
dataset |
metric |
score |
mindformers config |
---|---|---|---|---|
bert_base_uncased |
wiki |
- |
- |
|
bert_tiny_uncased |
wiki |
- |
- |
|
qa_bert_base_uncased |
SQuAD v1.1 |
EM / F1 |
80.74 / 88.33 |
|
tokcls_bert_base_chinese_cluener |
CLUENER |
Entity F1 |
0.7905 |
|
txtcls_bert_base_uncased_mnli |
Mnli |
Entity F1 |
84.80% |
|
clip_vit_b_32 |
Cifar100 |
Accuracy |
57.24% |
|
clip_vit_b_16 |
Cifar100 |
Accuracy |
61.41% |
|
clip_vit_l_14 |
Cifar100 |
Accuracy |
69.67% |
|
clip_vit_l_14@336 |
Cifar100 |
Accuracy |
68.19% |
|
glm_6b |
ADGEN |
BLEU-4 / Rouge-1 / Rouge-2 / Rouge-l |
8.42 / 31.75 / 7.98 / 25.28 |
|
gpt2 |
wikitext-2 |
- |
- |
|
gpt2_13b |
wikitext-2 |
- |
- |
|
gpt2_52b |
wikitext-2 |
- |
- |
|
llama_7b |
alpac |
- |
- |
|
llama_13b |
alpac |
- |
- |
|
mae_vit_base_p16 |
ImageNet-1K |
- |
- |
|
vit_base_p16 |
ImageNet-1K |
Accuracy |
83.71% |
|
pangualpha_2_6b |
悟道数据集 |
- |
- |
|
pangualpha_13b |
悟道数据集 |
- |
- |
|
swin_base_p4w7 |
ImageNet-1K |
Accuracy |
83.44% |
|
t5_small |
WMT16 |
- |
- |
|
t5_tiny |
WMT16 |
- |
- |
推荐
model |
dataset |
auc |
mindrec recipe |
vanilla mindspore |
---|---|---|---|---|
Wide&Deep |
Criteo |
0.8 |
||
Deep&Cross Network (DCN) |
Criteo |
0.8 |