We're thrilled to introduce the latest update, PandasAI v1.5. PandasAI is now faster than ever. We worked on optimizing it for speed and it's now about 2x cheaper (the prompt is 2x smaller) and at least 2x faster, depending on the query.
The biggest change is Pipelines. Pipelines leverage LLMs and are at the core of everything in PandasAI. Even the `.chat` method of the `SmartDataframe` and `SmartDatalake` now relies on Pipelines. We also added a new Pipeline to generate synthetic data with just a few samples. This can be used to train models, or for fake data in dev environments.
What's Changed
- New pipeline to create synthetic dataframes by ArslanSaleem (719)
- Use connector directly if flag is set by ArslanSaleem (731)
- Update prompt to construct dataframe using from_records by ArslanSaleem (741)
- Refactor Smart Data Frame Pipeline by milind-sinaptik (735)
- Update docs and examples for Azure OpenAI v1 by mspronesti (761)
- Rename sample_head to custom_head by gventuri
- Fix inconsistent charts in visualizations by gventuri
Breaking Changes
- Discontinued Falcon and StarCoder
- Discontinued callbacks
- Discontinued middlewares
New Contributors
- milind-sinaptik made their first contribution (735)
See the [full changelog](https://github.com/gventuri/pandas-ai/compare/v1.4.10...v1.5) for more details.
The key highlights are faster performance, lower cost, new synthetic data capabilities, and streamlined pipelines. Backwards incompatible changes were made by removing several features. Overall an exciting release that makes PandasAI more powerful and accessible!