📄 Hh Rlhf¶

Dataset Card for HH-RLHF Dataset Summary This repository provides access to two different kinds of data: Human preference data about helpfulness and harmlessness from Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback. These data are meant to train preference (or reward) models for subsequent RLHF training. These data are not meant for supervised training of dialogue agents. Training dialogue agents on these data is likely… See the full description on the dataset page: https://huggingface.co/datasets/Anthropic/hh-rlhf.

Tags: arxiv:2204.05862, croissant:True, human-feedback:True, license:mit, region:us

Note

ID: cards.hh_rlhf | Type: TaskCard

{
    "__description__": "Dataset Card for HH-RLHF\nDataset Summary\nThis repository provides access to two different kinds of data:\nHuman preference data about helpfulness and harmlessness from Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback. These data are meant to train preference (or reward) models for subsequent RLHF training. These data are not meant for supervised training of dialogue agents. Training dialogue agents on these data is likely… See the full description on the dataset page: https://huggingface.co/datasets/Anthropic/hh-rlhf.",
    "__tags__": {
        "arxiv": "2204.05862",
        "croissant": true,
        "human-feedback": true,
        "license": "mit",
        "region": "us"
    },
    "loader": {
        "path": "Anthropic/hh-rlhf",
        "type": "load_hf"
    },
    "preprocess_steps": [
        "splitters.small_no_dev",
        {
            "field": "chosen",
            "type": "strip"
        },
        {
            "field": "chosen",
            "new": "\n",
            "old": "\n\n",
            "type": "replace"
        },
        {
            "by": "\nAssistant:",
            "field": "chosen",
            "type": "split"
        },
        {
            "field": "chosen",
            "item": -1,
            "to_field": "output_choice",
            "type": "get"
        },
        {
            "field": "chosen",
            "stop": -1,
            "type": "slice"
        },
        {
            "by": "\nAssistant:",
            "field": "chosen",
            "to_field": "input",
            "type": "join"
        },
        {
            "by": "\nAssistant:",
            "field": "rejected",
            "type": "split"
        },
        {
            "field": "rejected",
            "item": -1,
            "to_field": "output_rejected",
            "type": "get"
        },
        {
            "fields": [
                "output_choice",
                "output_rejected"
            ],
            "to_field": "choices",
            "type": "list_field_values"
        },
        {
            "field": "choices",
            "type": "shuffle_field_values"
        },
        {
            "fields": {
                "input_type": "dialog",
                "instruction": "Respond the following dialog in an helpful and harmfull way.",
                "output_type": "response"
            },
            "type": "add_fields"
        }
    ],
    "task": "tasks.evaluation.preference",
    "templates": "templates.evaluation.preference.all",
    "type": "task_card"
}

Explanation about TaskCard¶

TaskCard delineates the phases in transforming the source dataset into a model-input, and specifies the metrics for evaluation of model-output.

Attributes:
loader: specifies the source address and the loading operator that can access that source and transform it into a unitxt multistream.

preprocess_steps: list of unitxt operators to process the data source into a model-input.

task: specifies the fields (of the already (pre)processed instance) making the inputs, the fields making the outputs, and the metrics to be used for evaluating the model output.

templates: format strings to be applied on the input fields (specified by the task) and the output fields. The template also carries the instructions and the list of postprocessing steps, to be applied to the model output.

Explanation about ListFieldValues¶

Concatenates values of multiple fields into a list, and assigns it to a new field.

Explanation about ShuffleFieldValues¶

Shuffles a list of values found in a field.

Explanation about AddFields¶

Adds specified fields to each instance in a given stream or all streams (default) If fields exist, updates them.

Args:

fields (Dict[str, object]): The fields to add to each instance.
Use ‘/’ to access inner fields

use_deepcopy (bool) : Deep copy the input value to avoid later modifications

Examples:
# Add a ‘classes’ field with a value of a list “positive” and “negative” to all streams AddFields(fields={“classes”: [“positive”,”negatives”]})

# Add a ‘start’ field under the ‘span’ field with a value of 0 to all streams AddFields(fields={“span/start”: 0}

# Add a ‘classes’ field with a value of a list “positive” and “negative” to ‘train’ stream AddFields(fields={“classes”: [“positive”,”negatives”], apply_to_stream=[“train”]})

# Add a ‘classes’ field on a given list, prevent modification of original list # from changing the instance. AddFields(fields={“classes”: alist}), use_deepcopy=True) # if now alist is modified, still the instances remain intact.

References: splitters.small_no_dev, templates.evaluation.preference.all, tasks.evaluation.preference