Data class for intermediate outputs of BLIP models. image_embeds (torch.FloatTensor): Image embeddings, shape (batch_size, num_patches, embe

Subclass torch's LayerNorm to handle fp16.

A class that keeps all conversation history.

Track a series of values and provide access to smoothed values over a window or the global series average.

Args: clip (torch.tensor, dtype=torch.uint8): Size is (C, T, H, W) Return: clip (torch.tensor, dtype=torch.float): Size is (T

Subclass torch's LayerNorm to handle fp16.

The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of cross-attention is added between

The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of cross-attention is added between

Data class of features from BlipFeatureExtractor. Args: image_embeds: (torch.FloatTensor) of shape (batch_size, num_patches+1, embed

Data class of features from AlbefFeatureExtractor. Args: image_embeds: `torch.FloatTensor` of shape `(batch_size, 1, embed_dim)`, `o

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`LlamaDecoderLayer`] Args: config: LlamaConfig

Image to Patch Embedding

Vision Transformere

Data class of features from AlbefFeatureExtractor. Args: image_embeds: `torch.FloatTensor` of shape `(batch_size, num_patches+1, emb

Attention based 2D feature pooling w/ learned (absolute) pos embedding. This is a multi-head attention based replacement for (spatial) average poo

Construct the embeddings from word and position embeddings.

Construct the embeddings from word and position embeddings.

Construct the embeddings from word and position embeddings.

The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of cross-attention is added between

Class for BLIP feature extractor. Supported model types: - base: BLIP base model with pre-trained weights from capfilt by BLIP large