Overwrite the `push_to_hub` method in order to force-add the tag "sft" when pushing the model on the Hub. Please refer to `~transform

r""" Save the pretrained model to a directory. This method is a wrapper around `transformers.PreTrainedModel.save_pretrained`. Please

Returns the 2d U-Net model used for diffusion.

Recursively unwraps a model from potential containers (as used in distributed training). Args: model (`torch.nn.Module`): The model

Return the directory part of a url.

A context manager to partition the model parameters during the model construction with MiCS partition strategy. Model states are partitioned

Do not add eos again if user already added it.

blocking call avoid explicit memory copy in _allgather_params

Helper that gets a list of the model's attention modules. Each module has a `bias` buffer used for causal masking. The Prefix LM conv

This is somewhat similar to `_load_state_dict_into_model`, but deals with a model that has some or all of its params on a `meta` device. It r

Change the default dtype and return the previous one. This is needed when wanting to instantiate the model under specific dtype.

Applies Rotary Position Embedding to the query and key tensors. Args: q (`torch.Tensor`): The query tensor. k (`torch.Tensor`): T

Llama tokenizer does satisfy `enc(a + b) = enc(a) + enc(b)`. It does ensure `enc(a + b) = enc(a) + enc(a + b)[len(enc(a)):]`.

Calculate the instruction level accuracy for given Subject results

Check that a list of shards is well-formed. This checks that the list is a list of dictionaries, and that each dictionary has a "url" and a "

Mark the history as completed.

Computes log probabilities of the reference model for a single padded batch of a DPO specific dataset.

Compute per token rewards from scores and KL-penalty. Args: scores (`torch.FloatTensor`): Scores from th

r""" return a callable function that takes in a "messages" dataset and returns a formatted dataset, based on the tokenizer apply chat template

Creates a static reference copy of a model. Note that model will be in `.eval()` mode. Args: model (`PreTrainedModelWrapper`): The m

Disable the redundant torch default initialization to accelerate model creation.

Free underlying storage of a parameter.

Generate responses for a list of histories.

Compute the DPO loss and other metrics for the given batch of inputs for train or test. CHANGE: 1. add sft loss 2. all gather metrics

Read an arbitrary file from the Tar archive. Args: name (str): File name to retrieve. Returns: io.BufferedReader

num_frames is the max number of frames the model can support. frame_count is the number of frames in the input video. max_fps is the max

Get the index (e.g. 2) from the prediction (e.g. 'C')

Get the index (e.g. 2) from the prediction (e.g. 'C')

Get the RingAttn implementation type.

Returns the first found floating dtype in `state_dict` if there is one, otherwise returns the first dtype.