Computes N-gram repetition penalty as described in Appendix C.2 of https://arxiv.org/abs/2502.03373. Reference implementation from: https://g

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

Format reward function specifically for code responses. Args: language: Programming language supported by E2B https://e2b.dev/docs/code-i

Registers a LightEval task configuration. - Core tasks can be added from this table: https://github.com/huggingface/lighteval/blob/main/src/light

Registers a LightEval task configuration. - Core tasks can be added from this table: https://github.com/huggingface/lighteval/blob/main/src/light

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

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

Get the tokenizer for the model.

Helper function for setting up Weights & Biases logging tools.

Reward function that checks if the completion is the same as the ground truth.

Load and preprocess a dataset from Parquet files, and filter out samples exceeding a specified length. Args: dataset_name (str): The

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

Format reward function specifically for code responses. Args: language: Programming language supported by E2B https://e2b.dev/docs/code-i

Computes N-gram repetition penalty as described in Appendix C.2 of https://arxiv.org/abs/2502.03373. Reference implementation from: https://g

Get the tokenizer for the model.

Helper function for setting up Weights & Biases logging tools.

Reward function that evaluates code snippets using the E2B code interpreter. Assumes the dataset contains a `verification_info` column with test

Basic string normalization.

pad a 2D tensors (e.g. responses, logprobs) in the last dim to max_seq_length. input shape: [bs, seq_length] output shape: [bs, max_seq_l

pad a 2D tensors (e.g. responses, logprobs) in the last dim to max_seq_length. input shape: [bs, seq_length] output shape: [bs, max_seq_l

pad a 2D tensors (e.g. responses, logprobs) in the last dim to max_seq_length. input shape: [bs, seq_length] output shape: [bs, max_seq_l

pad a 2D tensors (e.g. responses, logprobs) in the last dim to max_seq_length. input shape: [bs, seq_length] output shape: [bs, max_seq_l

Scores all test cases in a subtask. Args: client: PistonClient instance for executing code subtask: Dictionary containing su

Generate word-level n-grams from text.

Thread-safe function to write a result to JSONL file

Determines the status code based on the score and feedback message. Args: score: The numeric score (0.0 to 1.0) feedback: Th

Fix common compilation errors for IOI problems.

let qwen answer a batch of questions

Batch data into lists of length n. The last batch may be shorter.

Build ngram lookup for documents.