`iەxdZddlZddlmZddlmZddlmZm Z ddl Z ddl Z ddl m Z ddlmZdd lmZdd lmZdd lmZmZdd lmZmZdd lmZddlmZmZddlm Z m!Z!m"Z"m#Z#ddl$m%Z%m&Z&ddl'm(Z(m)Z)ddl*m+Z+e#j,e-Z.ee"dGdde Z/ee"dGdde Z0dZ1Gdde j2Z3Gdde j2Z4 dDd!e j2d"e j5d#e j5d$e j5d%e e j5d&e6d'e6fd(Z7Gd)d*e j2Z8Gd+d,e j2Z9Gd-d.e j2Z:Gd/d0e j2Z;Gd1d2e j2Z<Gd3d4eZ=Gd5d6e j2Z>e"Gd7d8eZ?e"Gd9d:e?Z@Gd;dd?e?ZBe"d@GdAdBe?ZCgdCZDdS)Ez,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<dS)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"/home/jaya/work/projects/VOICE-AGENT/VIET/agent-env/lib/python3.11/site-packages/transformers/models/videomae/modeling_videomae.pyrr*sp +/FHU& '...8