Reward function that scales based on completion length using a cosine schedule. Shorter correct solutions are rewarded more than longer ones.

Creates a draft of a model card using the information available to the `Trainer`. Args: model_name (`str` or `None`, *op

Creates a draft of a model card using the information available to the `Trainer`. Args: model_name (`str` or `None`, *op

Reward function that checks if the completion has a specific format.

Reward function that checks if the completion has a specific format.

Reward function that checks if the completion has a specific format.

Reward function that checks if the completion has a specific format.

Reward function that checks if the completion has a specific format.

Reward function that checks if the reasoning process is enclosed within <think> and </think> tags, while the final answer is enclosed within <answer>

Reward function that checks if the reasoning process is enclosed within <think> and </think> tags, while the final answer is enclosed within <answer>

init model on this single gpu and let it answer asign chunk of questions

Compute length-based rewards to discourage overthinking and promote token efficiency. Taken from the Kimi 1.5 tech report: https://arxiv.org/abs/

Event called at the end of an epoch.

Assumes the model is either prompted to emit \\boxed{answer} or does so automatically

Assumes the model is either prompted to emit \\boxed{answer} or does so automatically

Assumes the model is either prompted to emit \\boxed{answer} or does so automatically

r"""Reward function that checks for clear step-by-step reasoning. Regex pattern: Step \d+: - matches "Step 1:", "Step 2:", etc. ^\

reward function the penalizes repetitions ref implementation: https://github.com/eddycmu/demystify-long-cot/blob/release/openrlhf/ope

reward function the penalizes repetitions ref implementation: https://github.com/eddycmu/demystify-long-cot/blob/release/openrlhf/ope

results: {image_id: {label: [bboxes...] } } :return coco json format: {image_id: category_id:

Calculates the raw score for the subtask as the minimum score across all test results. Returns: float: The rounded minim

Scores multiple subtasks for a submission. Args: client: PistonClient instance for executing code subtasks: List of dictiona

Determines the overall status of the subtask based on the worst status among test results. Status priorities are ordered from worst t

Reward function that checks if we produce the desired number of think and answer tags associated with `format_reward()`. Adapted from: https://gi

Test accuracy_reward with a correct answer.

Test accuracy_reward with an incorrect answer.

Test code format reward with correct Python format.

Test cosine_scaled_reward with various cases.

Test code format reward with different programming languages.

Test len_reward with different length correct answers.

Test len_reward with different length incorrect answers.