Return an iterator over the keys of the cache.

Return an iterator over the keys of the cache.

Updates running moments from batch's moments computed across ranks

Return an iterator over the values of the cache.

This method forwards all its arguments to SiglipTokenizer's [`~PreTrainedTokenizer.decode`]. Please refer to the docstring of this me

This method forwards all its arguments to SiglipTokenizer's [`~PreTrainedTokenizer.batch_decode`]. Please refer to the docstring of t

Close the dataset.

r""" Return the state_dict of the pretrained model.

Generate response with the model given the query tensor. call the `generate` method of the model. Args: query_te

r""" Instantiates a new model from a pretrained model from `transformers`. The pretrained model is loaded using the `from_pretrained`

Returns the tokenizer used for tokenizing text inputs

Compute mean of tensor with a masked values.

Disable the redundant torch default initialization to accelerate model creation.

Preprocess an image or batch of images. Args: images (`ImageInput`): Image to preprocess. Expects a sing

`torch.device`: The device on which the module is (assuming that all the module parameters are on the same device).

Returns the model's input embeddings. Returns: `nn.Module`: A torch module mapping vocabulary to hidden states.

A method executed at the end of each Transformer model initialization, to execute code that needs the model's modules properly initia

Returns the autocast context manager

Extract frames from a video using OpenCV. Args: vpath_or_bytesio (str or BytesIO): Path to the video file or BytesIO object containi

r""" Post initialization method. This method is called after the model is instantiated and loaded from a checkpoint. It can be used to

Get the Ulysses sequence parallel process group.

r"""Loading the ZeRO-3/MiCS partitioned checkpoints Because the self.dp_process_group is replaced with the communicator for partition

Log `logs` on the various objects watching training. Subclass and override this method to inject custom behavior. Args:

r""" Instantiate a pretrained pytorch model from a pre-trained model configuration. The model is set in evaluation mode by default us

Ensure that numerator is divisible by the denominator.

Log `logs` on the various objects watching training, including stored metrics. Args: logs (`Dict[str, float]`):

Resize an image to `(size["height"], size["width"])`. Args: image (`np.ndarray`): Image to resize.

Tie the weights between the input embeddings and the output embeddings. If the `torchscript` flag is set in the configuration, can't

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

Run the given model on the given batch of inputs, concatenating the chosen and rejected inputs together. We do this to avoid doing two forwar

subclass can overload to specify different relative rank in parameter partition group