r"""Replace the jinja template in the tokenizer.

Fix common import and function changes between Python 3 versions Args: source_code (str): The Python source code to update Retu

r"""Add eos token and pad token to the tokenizer.

r"""Generate the assistant message including all the tool calls.

r"""Generate CLI commands for previewing.

r"""Plot loss curves in LlamaBoard.

Generate inference results and directly write to data chunks (like api_rewrite_documents). Args: model_path: Path to the model checkp

Generate inference results using vLLM engine (optimized for speed). Args: model_path: Path to the model checkpoint data_dir:

r"""Only applies gradient checkpointing to trainable layers.

r"""Get the attributes of the datasets.

Get default configuration for Researchy-GEO dataset. Returns: Configuration dictionary

Get elapsed time since start. Returns: Formatted elapsed time string

r"""Return an iterator over all elements with their names.

Generate prompt for LLM to explain document preference. Args: query: User query text document_a: First document text

Get explanation response from LLM for document preference. Args: query: User query text document_a: First document text

Extract rules from explanation text using LLM. Args: explanation_text: Explanation text from LLM winner: Winner document

vLLM enforces a constraint that the number of attention heads must be divisible by the number of GPUs and 64 must be divisible by the number of GPUs.

r"""Get id by element.

r"""Generate model's responses given queries.

r"""Return a logger with the specified name. It it not supposed to be accessed externally.

r"""Get logits processor that removes NaN and Inf logits.

Merge and process rules through hierarchical merging, filtering, and contradiction detection. Args: rules_list: List of rules to m

r"""Get the model path according to the model name.

Creates a MorphCloudExecutionClient instance using environment variables. Environment variables: MORPH_API_KEY: API key for MorphClo

Function to get model param counts from safetensors metadata or find patterns like 42m, 1.5b, 0.5m or products like 8x7b in a repo ID.

r"""Get the peak memory usage for the current device (in Bytes).

Factory function to get the appropriate code execution provider. Args: provider_type: Type of provider to use ("e2b", "morph") **

r"""Compute scores using given reward model. Both inputs and outputs are put on CPU.

r"""Get reward scores from the API server.

Generate prompt for extracting rules from explanation text. Args: explanation_text: Explanation text from LLM winner: Win

Generate prompt to filter rules, removing query-specific dependencies. Args: rule: Rule text that may contain query-specific refer

Get list of active piston worker endpoints from squeue output

r"""Get the callback for logging to SwanLab.

r"""Get the template name if the model is a chat/distill/instruct model.

r"""Get current date and time.

r"""Get training infomation for monitor. If do_train is True: Inputs: top.lang, train.output_path Outputs: train.output_box, trai

r"""Prepare the indices and seqlens for flash attn varlen function. Returns: indices: indices of non-masked tokens from the flattened seq

r"""Get the peft model for the pretrained model with unsloth. Used in training.

r"""Implement efficient greedy algorithm with binary search for the knapsack problem.

Construct training data source for GRPO. Args: num_examples: Number of examples to process data_dir: Directory containing dat

Construct evaluation data for GRPO training. Args: num_examples: Number of examples to process data_dir: Directory containing

r"""Check if the path has a tokenized dataset.

Construct inference data for model evaluation. Args: num_examples: Number of examples to process data_dir: Directory containi

Reward function that evaluates IOI problems using a specified execution client. Assumes the dataset has the same format as hf.co/datasets/open-r1

r"""Check if the accelerator is available.

r"""Judge if the model is a vision language model.

r"""Load dataset_info.json.

Load preference dataset from JSON file. Args: file_path: Path to JSON file containing preference data Returns:

r"""Get scores after evaluation.

Load IOI tests for a given year and problem id.

Load IOI tests for a given year.

r"""Load peft model with unsloth. Used in both training and inference.

r"""Optionally load pretrained model with unsloth. Used in training.

Main entry point for AutoGEO evaluation pipeline.