Generate predictions for a batch using the model. Args: batch: Input batch Returns: Dict[str, torch

Forward pass returning *continuous* latent phenotypes, with optional GE non‑linearity, followed by task‑specific linear head. Used by both GE

Predict structure and MFE for a single sequence using ViennaRNA with bp-span control.

Prepare a batch for model input. Args: batch: Input batch Returns: Prepared batch

Process labels for metric computation. Args: labels (torch.Tensor): Raw labels Returns: np.ndarray:

Process predictions for metric computation. Args: predictions (torch.Tensor): Raw predictions Returns:

Save model checkpoint after each epoch. Args: path_to_save (str): Base path for saving epoch (int): Current

Save the final trained model. Args: path_to_save (str): Base path for saving **kwargs: Additional keyword ar

Save model state dictionary to temporary file.

Save the current model state dictionary.

This method adds standard arguments that are common across all OmniGenome CLI commands, such as logging level and output directory.

Apply noise to a single sequence by randomly masking tokens. Args: seq (str): Input genomic sequence Returns:

This function parses command-line arguments, initializes the AutoBench, and runs the evaluation.

Check if benchmark version is compatible with OmniGenBench version. This function compares the benchmark version with the OmniGenBench versi

Clean up the generated Python script to make it testable. Removes interactive elements and notebook magic commands. Args: sc

Clean up temporary checkpoint files older than specified days. This function removes temporary checkpoint files that are older than the

Compute Base Pairing Probability matrix using ViennaRNA. This method uses the partition function approach to calculate the

3x3 convolution with padding. Args: in_planes (int): Number of input channels out_planes (int): Number of output channels

Convert a Jupyter notebook to a Python script using nbconvert. Args: notebook_path: Path to the .ipynb file Returns

Create a demo dataset for testing. This function generates synthetic RNA sequences with random labels for demonstration purposes. In

Evaluate the model on the validation dataset. Returns: Dict[str, Any]: Dictionary containing evaluation metrics

End‑to‑end training loop with a 60/20/20 train/val/test split. Returns the trained *nn.Module* and the sequence list (for external eva

Generate an SVG string of the RNA fold.

Extract base pairs from dot-bracket structure

Aggregate a list of numeric values. Args: result (list): A list of numeric values to aggregate Returns:

Default identity post-processing (override in subclasses).

Predicts the RNA structure for a single sequence.

This function creates a ResNet model with 16 basic blocks, optimized for processing genomic sequences and their structural representations.

Save the trained model. Args: path (str): Path to save the model overwrite (bool): Whether to overwrite exis

Save the trained model. Args: path_to_save (str): Path to save the model overwrite (bool): Whether to overwr

Save the trained model. Args: path_to_save (str): Path to save the model overwrite (bool): Whether to overwr

Convert SVG string to HTML code that can be displayed in Gradio HTML component.

Convert sequence(s) into k-mers. This method breaks the input sequence(s) into overlapping k-mers based on the configured k-

Train a translation efficiency prediction model with BPP features. This is a high-level function that handles the entire training pipeli

Initialize the ModelHub instance. Args: *args: Additional positional arguments **kwargs: Additional keyword

Initialize the OmniPooling layer. Args: config: Model configuration object containing architecture information

Initialize the HuggingFace trainer wrapper. Args: model (torch.nn.Module): The model to be trained training_

Initialize the dataset. Args: dataset_name_or_path: Path to dataset or dataset name tokenizer: Token

String representation of the cache.

Return a formatted, human-readable string representation of the metrics. Returns: str: A formatted string showing metric

Set the value of an argument and add it to the argument dict and call count dict. :param arg_name: The name of the argument.

Return a serializable dictionary for ``save_pretrained``. The dictionary includes both backbone hyperparameters and head/task info.

Internal method to calculate loss if not already present in outputs. :param outputs: The dictionary of model outputs. :param

Collect all metadata to be saved with the model. This method gathers comprehensive metadata about the model, including: - Lo

Compute loss from model outputs. Args: outputs (Dict[str, torch.Tensor]): Model outputs Returns: to

Compute loss from model outputs. Args: outputs (Dict[str, torch.Tensor]): Model outputs Returns: to

Multi-point crossover between randomly chosen parents.

Resolve the provided identifier to a local directory, downloading from Hugging Face first and falling back to the OmniGenome hub.

Evaluate sequences by folding and computing hamming distance to target structure; select via NSGA-like fronts.

Search for a class inside a package (OmniGenBench first).

Forward pass implementation. Args: x (Tensor): Input tensor [batch_size, channels, height, width] Returns:

(Private) Generates an in-silico MAVE dataset. This helper function creates a dataset of mutated sequences by randomly perturbing the

(Private) Generate single-mutation variants of the sequence. This function creates all possible point mutations of the input sequence,

(Private) Generates an in-silico MAVE dataset. This helper function creates a dataset of mutated sequences and uses the provided deep

Infer the optimization direction based on metric values. This function analyzes the trend of metric values to determine whether larger v

Initialize weights for custom layers using Xavier initialization.

Initialize weights for new layers.

Initialize the population via MLM conditioned on target structure.

Check if the model is a causal language model. Determines if the model architecture is causal based on the configuration. R

Load model configuration and check for differences with current config. This method loads the saved configuration and compares it wi

Restore the dataset class from metadata. This method attempts to dynamically import the dataset class that was saved with th

Load dataset_info.json from the dataset directory. This method searches for dataset_info.json in the directory containing th

Restore saved loss function from metadata. This method attempts to restore the loss function that was used during training b

Load and validate metadata from saved model directory. This method reads the metadata.json file and performs validation to e

Load model state dictionary from temporary file.

Load the best model state dictionary.

Load saved tokenizer from binary file. This method deserializes the tokenizer that was saved using dill. The tokenizer is es

Load model weights with validation and compatibility checks. This method loads the saved state dictionary and performs thorough

Create a layer of ResNet blocks. Args: block: Type of ResNet block to use planes (int): Number of output cha

(Private) Predicts the model's output for a list of mutated sequences. This helper function queries the model for the predictions of a list

(Private) Predict the model's output for a list of mutated sequences. This function uses the model to predict the output for each mutated seq

Pads and truncates sequences in the dataset to a uniform length. The length is determined dynamically based on the longest sequence i

Performs postprocessing on the tokenized data. This method standardizes the 'labels' field and prints the label distribution

Performs preprocessing on the loaded examples. This method standardizes the 'sequence' field and adds secondary structure inf

Remove temporary state dictionary file.

Remove the temporary model state dictionary file.

Resolve the model class, prioritizing OmniGenBench, then custom files.

Copy base model configuration files to save directory. This method copies essential configuration files from the original model

Save custom dataset class source file if available. This method saves the source code of custom dataset classes associated w

Save custom model class source file if it's user-defined. This method detects and saves the source code of custom model classes

Save model weights and tokenizer to disk. This method handles the serialization of model weights and tokenizer. It attempts