generic.py (1041B)
1 dataname = "all" 2 model = "generic" 3 4 use_clip = True 5 modelname = 'ViT-L/14' #"ViT-B/16" 6 workers = 3 7 amp = False 8 unfreeze="classifier" 9 pooling = "mlp" 10 use_caption = False 11 use_entities = False 12 features_path = False #"features_b7" 13 use_sbert = False 14 sbert_model = '/gpfswork/rech/lco/url46ht/.cache/torch/sentence_transformers/sentence-transformers_paraphrase-distilroberta-base-v1/' 15 16 use_visualbert = False 17 visualbert_name = 'uclanlp/visualbert-vcr-coco-pre'#'uclanlp/visualbert-nlvr2-coco-pre' 18 visualbert_text_model = "bert-base-uncased" #limjiayi/bert-hateful-memes-expanded facebook/bart-large 19 20 from configs.label_distribution import label_frequencies 21 from configs.micro_to_macro import multilabel, batch_per_epoch, cyclic_subsample, subsample_labels 22 23 # Model parameters 24 hidden_size = 768 25 dropout = 0 26 27 # Optimizer parameters 28 learning_rate = 1e-4 29 weight_decay = 0.001 30 max_grad_norm = 1 31 32 # Trainer parameters 33 batch_size = 8 34 accumulation = 16 35 max_epoch = 25 36 patience = 3 37 38 max_text_len = 128 39 max_nb_ent = 10 40 max_ent_len = 15 41 image_max_edge = 384