CrossCodeEval: A Diverse and Multilingual Benchmark for Cross-File Code Completion

This repository contains the data and inference code of the NeurIPS 2023 (Datasets and Benchmarks track) paper "CrossCodeEval: A Diverse and Multilingual Benchmark for Cross-File Code Completion."

Requirements

• Uncompress the CrossCodeEval data via tar -xvJf data/crosscodeeval_data.tar.xz -C data/
  • The data contains {baseline, retrieval, retrieval w/ ref.} setting x {bm25, UniXCoder, OpenAI Ada} retriever.
  • Please email us if you need the raw data.
• Install dependencies via pip install -r requirements.txt
• Build tree sitter via bash scripts/build_treesitter.sh

Evaluation on CrossCodeEval

Our evaluation consists of two steps: generation and metrics calculation.

Generation

Publicly Available Models

For publicly available models like StarCoder, DeepSeek-Coder, etc., we recommended using vLLM for fast and distributed inference on CrossCodeEval.

bash export gpus=2 export model=bigcode/starcoder2-3b export language=python export task=line_completion_rg1_unixcoder_cosine_sim export output_dir=./tmp/crosscodeeval_testrun/ python scripts/vllm_inference.py \ --tp $gpus \ --task $task \ --language $language \ --model $model \ --output_dir $output_dir \ --use_crossfile_context For additional args, e.g., cross-file context length and sampling topp, please see `python vllminference.py --help`.

OpenAI models

OpenAI models are accessible through an API. You may use the following script: ```bash export model=gpt-3.5-turbo-0125 export language=python export task=linecompletionrg1unixcodercosinesim export outputdir=./tmp/crosscodeevalopenaitestrun/ python scripts/openaiinference.py \ --task $task \ --language $language \ --model $model \ --outputdir $outputdir \ --usecrossfile_context

```

Metrics Calculation

After obtaining the generation, we can calculate the final metrics bash export language=python export ts_lib=./build/${language}-lang-parser.so; export task=line_completion_oracle_unixcoder_cosine_sim export prompt_file=./data/${language}/${task}.jsonl export output_dir=./tmp/crosscodeeval_testrun/; python scripts/eval.py \ --prompt_file $prompt_file \ --output_dir $output_dir \ --ts_lib $ts_lib \ --language $language \ --only_compute_metric

Citation

```

@inproceedings{ding2023crosscodeeval, title={CrossCodeEval: A Diverse and Multilingual Benchmark for Cross-File Code Completion}, author={Yangruibo Ding and Zijian Wang and Wasi Uddin Ahmad and Hantian Ding and Ming Tan and Nihal Jain and Murali Krishna Ramanathan and Ramesh Nallapati and Parminder Bhatia and Dan Roth and Bing Xiang}, year={2023}, booktitle={Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track}, url={https://arxiv.org/pdf/2310.11248.pdf} } ```

Questions

Please feel free to email us. You may also submit an issue in this repo.

Security

See CONTRIBUTING for more information.

License

This project is licensed under the Apache-2.0 License.