h`dZddlZddlmZddlmZddlmZm Z ddl Z ddl Z ddl Z ddl mZddlmZdd lmZdd lmZdd lmZmZdd lmZmZdd lmZddlmZmZddl m!Z!m"Z"m#Z#m$Z$ddl%m&Z&m'Z'ddl(m)Z)m*Z*ddl+m,Z,e$jZe.Z/ee#dGdde!Z0ee#dGdde!Z1dZ2GddejfZ4GddejfZ5 dCd ejfd!e jld"e jld#e jld$e e jld%e7d&e7fd'Z8Gd(d)ejfZ9Gd*d+ejfZ:Gd,d-ejfZ;Gd.d/ejfZ<Gd0d1ejfZ=Gd2d3eZ>Gd4d5ejfZ?e#Gd6d7eZ@e#Gd8d9e@ZAGd:d;ejfZBe#d<Gd=d>e@ZCe#d?Gd@dAe@ZDgdBZEy)Dz,PyTorch VideoMAE (masked autoencoder) model.N)deepcopy) dataclass)CallableOptional)nn)MSELoss)ACT2FN)GradientCheckpointingLayer)BaseModelOutputImageClassifierOutput)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack) find_pruneable_heads_and_indicesprune_linear_layer) ModelOutputTransformersKwargsauto_docstringlogging)IMAGENET_DEFAULT_MEANIMAGENET_DEFAULT_STD)can_return_tuplecheck_model_inputs)VideoMAEConfigz[ Class for VideoMAEDecoder's outputs, with potential hidden states and attentions. ) custom_introceZdZUdZdZeejed<dZ ee ejed<dZ ee ejed<y)VideoMAEDecoderOutputz logits (`torch.FloatTensor` of shape `(batch_size, patch_size ** 2 * num_channels)`): Pixel reconstruction logits. Nlogits hidden_states attentions) __name__ __module__ __qualname____doc__r rtorch FloatTensor__annotations__r!tupler"l/var/www/html/eduruby.in/venv/lib/python3.12/site-packages/transformers/models/videomae/modeling_videomae.pyrr+sR +/FHU&& '.8