This function is the entry point for the 'autotrain' console script.

Returns a random sample of n items from the dataset. Args: n (int): The number of samples to return. Returns:

Sample benchmark configuration. Based on examples/autobench_gfm_evaluation/RGB/*/config.py pattern.

从测试集中抽样，返回序列与 Ground Truth 结构

Sample RNA sequence from test dataset

Save the LoRA-adapted model using the base model's save functionality. This method delegates saving operations to the base model whi

This method saves the complete pipeline including model, tokenizer, datasets, trainer, and metadata to a directory. The saved pipelin

This function saves the arguments from a configuration object to a text file. It's useful for logging experiment parameters and configuration

Save the generated embeddings to a file. Args: embeddings (torch.Tensor): The embeddings to save output_path

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

Save model weights, config, tokenizer and metadata to a directory. Files ----- - ``config.json``: Backbone/head configuration

Save backbone weights and config to ``save_directory``.

Save backbone weights and config to a directory.

Set a custom loss function for the base model. This method allows configuration of specialized loss functions through the ba

Optionally set the id->one-hot projection matrix of shape (vocab_size, 4).

Configure warnings for tests

Calculate validity loss for RNA structure

Create temporary output directory for model checkpoints

Create temporary test data directory

Test the model on the test dataset. Returns: Dict[str, Any]: Dictionary containing test metrics

Test that same aggregation gives consistent results

Verify all subcommands provide help

Test extracting all layers and all heads (most expensive operation)

Test attention statistics calculation. Based on section 1 of attention_extraction_example.py

Test that AutoBench has all expected attributes

Test that benchmark names are handled correctly

Test autobench accepts different benchmark names

Test that AutoBench accepts cache_dir parameter

Test autobench parser is properly configured

Test AutoBench accepts various configuration options

Test AutoBench can be initialized

Test AutoBench supports multi-seed evaluation

Test autobench accepts overwrite flag

Test autobench parameter mapping from ogb to bench_command

Test that parameter names in ogb_cli match those expected by bench_command

Test autobench requires model and benchmark

Test AutoBench.run() with minimal configuration. This is a slow test that actually runs benchmarking.

Test that AutoBench.run() has the correct signature

Test that short options (-m, -b, -t) work correctly

Test autobench accepts different trainer types

Test AutoBench accepts different benchmark names

Test that AutoConfig behaves like a dictionary

Test that AutoConfig has the from_hub classmethod

Test AutoConfig initialization from a config dict

Test AutoConfig.get() method with defaults

Test that AutoConfig parameters can be overridden

Test autoinfer accepts batch size parameter

Test autoinfer parser is properly configured

Test autoinfer validates input requirements

Test autoinfer parameter parsing

Test autoinfer requires model argument

Test autoinfer with single sequence input

Test that AutoTrain accepts cache_dir parameter for local datasets

Test autotrain parser is properly configured

Test that dataset names are handled correctly for metrics

Test AutoTrain handles different task types

Test autotrain command execution with mock dataset

Test complete AutoTrain workflow. This is a slow integration test.

Test AutoTrain can be initialized

Test autotrain accepts optional training parameters

Test autotrain parameter mapping from ogb to train_command

Test that parameter names in ogb_cli match those expected by train_command

Test autotrain requires dataset and model

Test that AutoTrain.run() accepts a config parameter

Test AutoTrain accepts custom training configuration

Test that local benchmark workflows still work

Test that local dataset workflows still work (backward compatibility)

Test counting base pairs in structures

Test calculating precision and recall for base pairs

Test basic encoding with mean aggregation

Test complete training workflow for binary classification. Based on quickstart_te.py pattern. Uses synthetic data with minima

Test batch extraction of attention scores. Based on section 2 of attention_extraction_example.py

Test batch design of multiple structures. Based on example_3_batch_design() from rna_design_examples.py

Test batch processing with different batch sizes

Test efficient batch processing. Based on batch example in README.md

Test batch inference using loop. Based on Python API batch example in README.md

Test processing multiple sequences

Test batch structure prediction

Test that batch encoding gives same results as individual

Test batch processing with sequences of different lengths

Test benchmarking multiple models in sequence

Test benchmark results follow expected format

Test benchmark datasets have standard splits

Test that boolean flags use action='store_true' consistently

Test BPP matrix computation.

Test BPP processor module.

Create a temporary cache directory for testing.

Test classification metrics for structure prediction

Test classification metrics are available

Test metrics with ignore index (-100)

Test attention extraction from classification model