|
|
|
|
|
|
|
|
|
|
|
model:
|
|
arch: albef_retrieval
|
|
load_finetuned: True
|
|
|
|
pretrained: "https://storage.googleapis.com/sfr-pcl-data-research/ALBEF/ALBEF.pth"
|
|
finetuned: "https://storage.googleapis.com/sfr-vision-language-research/LAVIS/models/ALBEF/albef_coco_retrieval_lavis.pt"
|
|
|
|
queue_size: 65536
|
|
|
|
|
|
vit_type: "base"
|
|
image_size: 384
|
|
vit_ckpt_layer: 0
|
|
vit_drop_path_rate: 0
|
|
vit_layer_norm_epsilon: 1e-6
|
|
vit_grad_ckpt: False
|
|
|
|
|
|
med_config_path: "configs/models/med_config_albef.json"
|
|
|
|
embed_dim: 256
|
|
momentum: 0.995
|
|
alpha: 0.4
|
|
temp: 0.07
|
|
use_distill: True
|
|
|
|
max_txt_len: 30
|
|
|
|
preprocess:
|
|
vis_processor:
|
|
train:
|
|
name: "blip_image_train"
|
|
image_size: 384
|
|
eval:
|
|
name: "blip_image_eval"
|
|
image_size: 384
|
|
text_processor:
|
|
train:
|
|
name: "blip_caption"
|
|
eval:
|
|
name: "blip_caption"
|
|
|