SHARK-Studio/tank/model_metadata.csv

1	model_name	use_tracing	dynamic	param_count	tags	notes
2	microsoft/MiniLM-L12-H384-uncased	True	True	66M	nlp;bert-variant;transformer-encoder	Large version has 12 layers; 384 hidden size; Smaller than BERTbase (66M params vs 109M params)
3	albert-base-v2	True	True	11M	nlp;bert-variant;transformer-encoder	12 layers; 128 embedding dim; 768 hidden dim; 12 attention heads; Smaller than BERTbase (11M params vs 109M params); Uses weight sharing to reduce # params but computational cost is similar to BERT.
4	bert-base-uncased	True	True	109M	nlp;bert-variant;transformer-encoder	12 layers; 768 hidden; 12 attention heads
5	bert-base-uncased_fp16	True	True	109M	nlp;bert-variant;transformer-encoder	12 layers; 768 hidden; 12 attention heads
6	bert-base-cased	True	True	109M	nlp;bert-variant;transformer-encoder	12 layers; 768 hidden; 12 attention heads
7	distilbert-base-uncased	True	True	66M	nlp;bert-variant;transformer-encoder	Smaller and faster than BERT with 97percent retained accuracy.
8	google/mobilebert-uncased	True	True	25M	nlp,bert-variant,transformer-encoder,mobile	24 layers, 512 hidden size, 128 embedding
9	alexnet	False	True	61M	cnn,parallel-layers	The CNN that revolutionized computer vision (move away from hand-crafted features to neural networks),10 years old now and probably no longer used in prod.
10	resnet18	False	True	11M	cnn,image-classification,residuals,resnet-variant	1 7x7 conv2d and the rest are 3x3 conv2d
11	resnet50	False	True	23M	cnn,image-classification,residuals,resnet-variant	Bottlenecks with only conv2d (1x1 conv -> 3x3 conv -> 1x1 conv blocks)
12	resnet50_fp16	False	True	23M	cnn,image-classification,residuals,resnet-variant	Bottlenecks with only conv2d (1x1 conv -> 3x3 conv -> 1x1 conv blocks)
13	resnet101	False	True	29M	cnn,image-classification,residuals,resnet-variant	Bottlenecks with only conv2d (1x1 conv -> 3x3 conv -> 1x1 conv blocks)
14	squeezenet1_0	False	True	1.25M	cnn,image-classification,mobile,parallel-layers	Parallel conv2d (1x1 conv to compress -> (3x3 expand | 1x1 expand) -> concat)
15	wide_resnet50_2	False	True	69M	cnn,image-classification,residuals,resnet-variant	Resnet variant where model depth is decreased and width is increased.
16	mobilenet_v3_small	False	True	2.5M	image-classification,cnn,mobile	N/A
17	google/vit-base-patch16-224	True	False	86M	image-classification,vision-transformer,transformer-encoder	N/A
18	microsoft/resnet-50	True	False	23M	image-classification,cnn,residuals,resnet-variant	Bottlenecks with only conv2d (1x1 conv -> 3x3 conv -> 1x1 conv blocks)
19	facebook/deit-small-distilled-patch16-224	True	False	22M	image-classification,vision-transformer,cnn	N/A
20	microsoft/beit-base-patch16-224-pt22k-ft22k	True	False	86M	image-classification,transformer-encoder,bert-variant,vision-transformer	N/A
21	nvidia/mit-b0	True	False	3.7M	image-classification,transformer-encoder	SegFormer
22	camembert-base	False	False	-	-	-
23	dbmdz/convbert-base-turkish-cased	False	False	-	-	-
24	google/electra-small-discriminator	False	False	-	-	-
25	hf-internal-testing/tiny-random-flaubert	False	False	-	-	-
26	funnel-transformer/small	False	False	-	-	-
27	microsoft/layoutlm-base-uncased	False	False	-	-	-
28	microsoft/mpnet-base	False	False	-	-	-
29	roberta-base	False	False	-	-	-
30	xlm-roberta-base	False	False	-	-	-
31	facebook/convnext-tiny-224	False	False	-	-	-
32	efficientnet-v2-s	False	False	22M	image-classification,cnn	Includes MBConv and Fused-MBConv
33	mnasnet1_0	False	True	-	cnn, torchvision, mobile, architecture-search	Outperforms other mobile CNNs on Accuracy vs. Latency
34	bert-large-uncased	True	True	330M	nlp;bert-variant;transformer-encoder	24 layers, 1024 hidden units, 16 attention heads
35	t5-base	True	False	220M	nlp;transformer-encoder;transformer-decoder	Text-to-Text Transfer Transformer
36	t5-large	True	False	770M	nlp;transformer-encoder;transformer-decoder	Text-to-Text Transfer Transformer
37	bert-large-uncased	True	hf	True	330M	nlp;bert-variant;transformer-encoder	24 layers, 1024 hidden units, 16 attention heads
38	efficientnet_b0	True	False	5.3M	image-classification;cnn;conv2d;depthwise-conv	Smallest EfficientNet variant with 224x224 input
39	efficientnet_b7	True	False	66M	image-classification;cnn;conv2d;depthwise-conv	Largest EfficientNet variant with 600x600 input
40	gpt2	True	False	110M	nlp;transformer-decoder;auto-regressive	12 layers, 768 hidden units, 12 attention heads
41	t5-base	True	False	220M	nlp;transformer-encoder;transformer-decoder	Text-to-Text Transfer Transformer
42	t5-large	True	False	770M	nlp;transformer-encoder;transformer-decoder	Text-to-Text Transfer Transformer