Source code for nemo_rl.data.hf_datasets.helpsteer3

# Copyright (c) 2025, NVIDIA CORPORATION.  All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from absl import logging
from datasets import load_dataset

from nemo_rl.data.interfaces import TaskDataSpec




def format_helpsteer3(data):
    response_1 = data["response1"]
    response_2 = data["response2"]
    overall_preference = data["overall_preference"]

    if overall_preference < 0:
        chosen = response_1
        rejected = response_2
    elif overall_preference == 0:
        logging.log_every_n(
            logging.WARNING,
            "Preference is 0 for some examples! Setting chosen and rejected to response 1 since we don't know which response is better",
            1000,
        )
        chosen = response_1
        rejected = response_1
    else:
        chosen = response_2
        rejected = response_1

    return {
        "prompt": data["context"],
        "chosen_response": chosen,
        "rejected_response": rejected,
    }





class HelpSteer3Dataset:
    """HelpSteer3 preference dataset for DPO training."""

    def __init__(self):
        ds = load_dataset("nvidia/HelpSteer3", "preference")
        self.formatted_ds = ds.map(format_helpsteer3)

        self.task_spec = TaskDataSpec(
            task_name="HelpSteer3",
        )