README.md

---
dataset_info:
- config_name: librusec_history
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: 'null'
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: string
  splits:
  - name: test
    num_bytes: 9288308
    num_examples: 128
  download_size: 287104
  dataset_size: 9288308
- config_name: librusec_mhqa
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: 'null'
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: 'null'
  splits:
  - name: test
    num_bytes: 6973356
    num_examples: 384
  download_size: 339262
  dataset_size: 6973356
- config_name: long_context_multiq
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: 'null'
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: string
  splits:
  - name: test
    num_bytes: 137088574
    num_examples: 1200
  download_size: 65723227
  dataset_size: 137088574
- config_name: matreshka_names
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: 'null'
  - name: metadata
    struct:
    - name: id
      dtype: int64
    - name: source_path
      dtype: string
  splits:
  - name: test
    num_bytes: 110197693
    num_examples: 900
  download_size: 10212227
  dataset_size: 110197693
- config_name: matreshka_yes_no
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: string
  - name: metadata
    struct:
    - name: id
      dtype: int64
    - name: source_path
      dtype: string
  splits:
  - name: test
    num_bytes: 220381278
    num_examples: 1799
  download_size: 10269004
  dataset_size: 220381278
- config_name: passkey
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: 'null'
  - name: metadata
    struct:
    - name: id
      dtype: int64
    - name: source_path
      dtype: string
  splits:
  - name: test
    num_bytes: 135062614
    num_examples: 1200
  download_size: 6706454
  dataset_size: 135062614
- config_name: passkey_with_librusec
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: 'null'
  - name: metadata
    struct:
    - name: id
      dtype: int64
    - name: source_path
      dtype: string
  splits:
  - name: test
    num_bytes: 138752007
    num_examples: 1200
  download_size: 68813259
  dataset_size: 138752007
- config_name: ru_2wikimultihopqa
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: 'null'
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: string
  splits:
  - name: test
    num_bytes: 20384223
    num_examples: 300
  download_size: 10115092
  dataset_size: 20384223
- config_name: ru_babilong_qa1
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: string
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: 'null'
  splits:
  - name: test
    num_bytes: 98592899
    num_examples: 600
  download_size: 50184619
  dataset_size: 98592899
- config_name: ru_babilong_qa2
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: string
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: 'null'
  splits:
  - name: test
    num_bytes: 98616350
    num_examples: 600
  download_size: 50202521
  dataset_size: 98616350
- config_name: ru_babilong_qa3
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: string
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: 'null'
  splits:
  - name: test
    num_bytes: 98734535
    num_examples: 600
  download_size: 50147272
  dataset_size: 98734535
- config_name: ru_babilong_qa4
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: string
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: 'null'
  splits:
  - name: test
    num_bytes: 98509371
    num_examples: 600
  download_size: 50170814
  dataset_size: 98509371
- config_name: ru_babilong_qa5
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: string
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: 'null'
  splits:
  - name: test
    num_bytes: 98450820
    num_examples: 600
  download_size: 50175383
  dataset_size: 98450820
- config_name: ru_gsm100
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: 'null'
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: string
  splits:
  - name: test
    num_bytes: 3094379
    num_examples: 100
  download_size: 43906
  dataset_size: 3094379
- config_name: ru_qasper
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: 'null'
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: string
  splits:
  - name: test
    num_bytes: 10664041
    num_examples: 203
  download_size: 2978374
  dataset_size: 10664041
- config_name: ru_quality
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: string
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: 'null'
  splits:
  - name: test
    num_bytes: 9095705
    num_examples: 202
  download_size: 427250
  dataset_size: 9095705
- config_name: ru_sci_abstract_retrieval
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: string
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: string
  splits:
  - name: test
    num_bytes: 270306154
    num_examples: 1240
  download_size: 111456155
  dataset_size: 270306154
- config_name: ru_sci_fi
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: string
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: 'null'
  splits:
  - name: test
    num_bytes: 7217269
    num_examples: 64
  download_size: 399160
  dataset_size: 7217269
- config_name: ru_sci_passage_count
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: 'null'
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: string
  splits:
  - name: test
    num_bytes: 110232625
    num_examples: 600
  download_size: 29891223
  dataset_size: 110232625
- config_name: ru_tpo
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: string
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: 'null'
  splits:
  - name: test
    num_bytes: 7921499
    num_examples: 251
  download_size: 284509
  dataset_size: 7921499
- config_name: ru_trec
  features:
  - name: length
    dtype: string
  - name: context
    dtype: string
  - name: input
    dtype: string
  - name: positive_outputs
    sequence: string
  - name: negative_outputs
    sequence: 'null'
  - name: metadata
    struct:
    - name: id
      dtype: string
    - name: source_path
      dtype: string
  splits:
  - name: test
    num_bytes: 18657206
    num_examples: 300
  download_size: 7280413
  dataset_size: 18657206
configs:
- config_name: librusec_history
  data_files:
  - split: test
    path: librusec_history/test-*
- config_name: librusec_mhqa
  data_files:
  - split: test
    path: librusec_mhqa/test-*
- config_name: long_context_multiq
  data_files:
  - split: test
    path: long_context_multiq/test-*
- config_name: matreshka_names
  data_files:
  - split: test
    path: matreshka_names/test-*
- config_name: matreshka_yes_no
  data_files:
  - split: test
    path: matreshka_yes_no/test-*
- config_name: passkey
  data_files:
  - split: test
    path: passkey/test-*
- config_name: passkey_with_librusec
  data_files:
  - split: test
    path: passkey_with_librusec/test-*
- config_name: ru_2wikimultihopqa
  data_files:
  - split: test
    path: ru_2wikimultihopqa/test-*
- config_name: ru_babilong_qa1
  data_files:
  - split: test
    path: ru_babilong_qa1/test-*
- config_name: ru_babilong_qa2
  data_files:
  - split: test
    path: ru_babilong_qa2/test-*
- config_name: ru_babilong_qa3
  data_files:
  - split: test
    path: ru_babilong_qa3/test-*
- config_name: ru_babilong_qa4
  data_files:
  - split: test
    path: ru_babilong_qa4/test-*
- config_name: ru_babilong_qa5
  data_files:
  - split: test
    path: ru_babilong_qa5/test-*
- config_name: ru_gsm100
  data_files:
  - split: test
    path: ru_gsm100/test-*
- config_name: ru_qasper
  data_files:
  - split: test
    path: ru_qasper/test-*
- config_name: ru_quality
  data_files:
  - split: test
    path: ru_quality/test-*
- config_name: ru_sci_abstract_retrieval
  data_files:
  - split: test
    path: ru_sci_abstract_retrieval/test-*
- config_name: ru_sci_fi
  data_files:
  - split: test
    path: ru_sci_fi/test-*
- config_name: ru_sci_passage_count
  data_files:
  - split: test
    path: ru_sci_passage_count/test-*
- config_name: ru_tpo
  data_files:
  - split: test
    path: ru_tpo/test-*
- config_name: ru_trec
  data_files:
  - split: test
    path: ru_trec/test-*
license: mit
language:
- ru
tags:
- nlp
- benchmark
- russian
- long-context
- evaluation
---

# LIBRA: Long Input Benchmark for Russian Analysis

<img src="logo.png" width="800" />

## Dataset Summary

LIBRA (Long Input Benchmark for Russian Analysis) is designed to evaluate the capabilities of large language models (LLMs) in understanding and processing long texts in Russian. This benchmark includes 21 datasets adapted for different tasks and complexities. The tasks are divided into four complexity groups and allow evaluation across various context lengths ranging from 4k up to 128k tokens.

## Dataset Structure

The datasets are divided into subsets based on context lengths: 4k, 8k, 16k, 32k, 64k, and 128k tokens. Each subset contains a different number of samples depending on the task complexity.

## Tasks and Complexity Groups

<img src="table.png" width="800" />

### Group I: Simple Information Retrieval
- **Passkey**: Extract a relevant piece of code number from a long text fragment. Based on the original [PassKey test](https://github.com/CStanKonrad/long_llama/blob/main/examples/passkey.py) from the m LongLLaMA’s GitHub repo.
- **PasskeyWithLibrusec**: Similar to Passkey but with added noise from Librusec texts.

### Group II: Question Answering and Multiple Choice
- **MatreshkaNames**: Identify the person in dialogues based on the discussed topic. We used [Matreshka](https://huggingface.co/datasets/zjkarina/matreshka) dataset and [Russian Names](https://www.kaggle.com/datasets/rai220/russian-cyrillic-names-and-sex/data) dataset to create this and the next task.
- **MatreshkaYesNo**: Indicate whether a specific topic was mentioned in the dialog.
- **LibrusecHistory**: Answer questions based on historical texts. Ideologically similiar to the [PassageRetrieval dataset](https://huggingface.co/datasets/THUDM/LongBench/viewer/passage_retrieval_en) from LongBench.
- **ruTREC**: Few-shot in-context learning for topic classification. Created by translating the [TREC dataset](https://huggingface.co/datasets/THUDM/LongBench/viewer/trec_e) from LongBench.
- **ruSciFi**: Answer true/false based on context and general world knowledge. Translation of [SciFi dataset](https://huggingface.co/datasets/L4NLP/LEval/viewer/sci_f) from L-Eval which originally was based on [SF-Gram](https://github.com/nschaetti/SFGram-dataset).
- **ruSciAbstractRetrieval**: Retrieve relevant paragraphs from scientific abstracts.
- **ruTPO**: Multiple-choice questions similar to TOEFL exams. Translation of the [TPO dataset](https://huggingface.co/datasets/L4NLP/LEval/viewer/tpo) from L-Eval.
- **ruQuALITY**: Multiple-choice QA tasks based on detailed texts. Created by translating the [QuALITY dataset](https://huggingface.co/datasets/L4NLP/LEval/viewer/quality) from L-Eval.

### Group III: Multi-hop Question Answering
- **ruBABILongQA**: 5 long-context reasoning tasks for QA using facts hidden among irrelevant information.
- **LongContextMultiQ**: Multi-hop QA based on Wikidata and Wikipedia.
- **LibrusecMHQA**: Multi-hop QA requiring information distributed across several text parts.
- **ru2WikiMultihopQA**: Translation of the [2WikiMultihopQA dataset](https://huggingface.co/datasets/THUDM/LongBench/viewer/2wikimqa_e) from LongBench.

### Group IV: Complex Reasoning and Mathematical Problems
- **ruSciPassageCount**: Count unique paragraphs in a long text. Uses the basic idea of the original [PassageCount dataset](https://huggingface.co/datasets/THUDM/LongBench/viewer/passage_count) from LongBench.
- **ruQasper**: Question Answering over academic research papers. Created by translating the [Qasper dataset](https://huggingface.co/datasets/THUDM/LongBench/viewer/qasper_e) from LongBench.
- **ruGSM100**: Solve math problems using Chain-of-Thought reasoning. Created by translating the [GSM100](https://huggingface.co/datasets/L4NLP/LEval/viewer/gsm100) dataset from L-Eval.


## Metrics

We use **Exact Match (EM)** and **F1** metrics in our tests. **EM** metric is used to evaluate the accuracy of the model's responses by comparing the predicted answers to the ground truth. It is particularly effective for tasks where precise matching of responses is critical, such as question answering and retrieval tasks.

<img src="table2.png" width="800" />

## Citation

```
@misc{churin2024longinputbenchmarkrussian,
      title={Long Input Benchmark for Russian Analysis}, 
      author={Igor Churin and Murat Apishev and Maria Tikhonova and Denis Shevelev and Aydar Bulatov and Yuri Kuratov and Sergei Averkiev and Alena Fenogenova},
      year={2024},
      eprint={2408.02439},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2408.02439}, 
}
```

## GitHub

For more details and code, please visit our [GitHub repository](https://github.com/ai-forever/LIBRA/).