Load the dataset and return a tuple: (splits_dict, answer_type). splits_dict: A dictionary where each key is a split name (e.g., 'tr

Load VQA dataset from parquet files

Compute variance of tensor with masked values.

Try to register HF model configuration class to serialize by value If trust_remote_code is set, and the model's config file specifies an

Pad a DataProto to size divisible by size_divisor Args: data (DataProto): the unpadded DataProto size_divisor (int): size divisor

Provide backwards compatibility for RoPE.

Print the model size.

Process image for vLLM multi-modal input, following ImageProcessMixin pattern. Args: image: PIL.Image.Image or dict with key "bytes"

Consolidates and grades vLLM outputs for a single question, returning a result dictionary.

Remove the obsolete checkpoints that exceed the save_limit.

Note that this operation is in-place

Sort the placement groups by node ip, all bundles in a single placement group should be on the same node. FSDPCheckpointManager saves sharde

spawn to a dictionary of worker groups, each with a subset of method with prefix.

Serializes this instance to a Python dictionary. Returns: Dict[str, Any]: Dictionary of all the attributes that make up

Removes all attributes from config which correspond to the default config attributes for better readability and serializes to a Pytho

Save this instance to a JSON file. Args: json_file_path (`str` or `os.PathLike`): Path to the JSON file

Serializes this instance to a JSON formatted string. Returns: str: JSON formatted string representing the configuration

Union two tensordicts.

Union two dict. Will throw an error if there is an item not the same object with the same key.

Update the policy with an iterator of DataProto Args: data (DataProto): an iterator over the DataProto that returns by

AnyPrecisionAdamW: a flexible precision AdamW optimizer with optional Kahan summation for high precision weight updates. Allo