This repository is for the analysis done in BigCode Project. You can find analysis of datasets, models, architecture choices and more.
-
Data analysis: In the folder
data_analysis, we provide code for data analysis:- Near deduplication
- Python data analysis:
- Natural language distribution in comments/docstrings
- Data decontamination for HumanEval and MBPP benchmarks
- Percentage of files that can be successfully compiled
- Percentage of configuration and test files Some notebooks with some early data and model loss analysis.
-
Multi-Query Attention experiments, for details please to multi_query_experiments/README.md)