finetuner.models module

class finetuner.models.MLP(input_size, hidden_sizes, bias=True, activation=None, l2=False)

Bases: finetuner.models._ModelStub

MLP model stub.

Parameters

• input_size (int) – Size of the input representations.

• hidden_sizes (List[int]) – A list of sizes of the hidden layers. The last hidden size is the output size.

• bias (bool) – Whether to add bias to each layer.

• activation (Optional[str]) – A string to configure activation function, relu, tanh or sigmoid. Set to None for no activation.

• l2 (bool) – Apply L2 normalization at the output layer.

name: str = 'mlp'

description: str = 'Simple MLP encoder trained from scratch'

task: str = 'any'

output_dim: Optional[int] = '-'

architecture: str = 'MLP'

options: Dict[str, Any]

class finetuner.models.ResNet50

Bases: finetuner.models._ModelStub

ResNet50 model stub.

name: str = 'resnet50'

description: str = 'Pretrained on ImageNet'

task: str = 'image-to-image'

output_dim: Optional[int] = '2048'

architecture: str = 'CNN'

options: Dict[str, Any]

class finetuner.models.ResNet152

Bases: finetuner.models._ModelStub

ResNet152 model stub.

name: str = 'resnet152'

description: str = 'Pretrained on ImageNet'

task: str = 'image-to-image'

output_dim: Optional[int] = '2048'

architecture: str = 'CNN'

options: Dict[str, Any]

class finetuner.models.EfficientNetB0

Bases: finetuner.models._ModelStub

EfficientNetB0 model stub.

name: str = 'efficientnet_b0'

description: str = 'Pretrained on ImageNet'

task: str = 'image-to-image'

output_dim: Optional[int] = '1280'

architecture: str = 'CNN'

options: Dict[str, Any]

class finetuner.models.EfficientNetB4

Bases: finetuner.models._ModelStub

EfficientNetB4 model stub.

name: str = 'efficientnet_b4'

description: str = 'Pretrained on ImageNet'

task: str = 'image-to-image'

output_dim: Optional[int] = '1280'

architecture: str = 'CNN'

options: Dict[str, Any]

class finetuner.models.OpenAICLIP

Bases: finetuner.models._ModelStub

OpenAICLIP model stub.

name: str = 'openai/clip-vit-base-patch32'

description: str = 'Pretrained on text image pairs by OpenAI'

task: str = 'text-to-image'

output_dim: Optional[int] = '768'

architecture: str = 'transformer'

options: Dict[str, Any]

class finetuner.models.BERT

Bases: finetuner.models._ModelStub

BERT model stub.

name: str = 'bert-base-cased'

description: str = 'Pretrained on BookCorpus and English Wikipedia'

task: str = 'text-to-text'

output_dim: Optional[int] = '768'

architecture: str = 'transformer'

options: Dict[str, Any]

class finetuner.models.SentenceTransformer

Bases: finetuner.models._ModelStub

SentenceTransformer model stub.

name: str = 'sentence-transformers/msmarco-distilbert-base-v3'

description: str = 'Pretrained BERT, fine-tuned on MS Marco'

task: str = 'text-to-text'

output_dim: Optional[int] = '768'

architecture: str = 'transformer'

options: Dict[str, Any]