SQLCoder slightly outperforms gpt-3.5-turbo for natural language to SQL generation tasks on our [sql-eval](https://github.com/defog-ai/sql-eval) framework, and significantly outperforms all popular open-source models. SQLCoder is fine-tuned on a base StarCoder model.
## Results
| model | perc_correct |
|-|-|
| gpt-4 | 74.3 |
| defog-sql-coder | 64.6 |
| gpt-3.5-turbo | 60.6 |
| defog-easy | 57.1 |
| text-davinci-003 | 54.3 |
| wizardcoder | 52.0 |
| starcoder | 45.1 |
## Training
Defog was trained on 10,537 human-curated questions across 2 epochs. These questions were based on 10 different schemas. None of the schemas in the training data were included in our evaluation framework.
Training happened in 2 phases. The first phase was on questions that were classified as "easy" or "medium" difficulty, and the second phase was on questions that were classified as "hard" or "extra hard" difficulty. You can read more about the dataset creation and classification process [here](https://defog.ai/blog/defog-sql-coder-dataset-creation).
## Results by question category
We classified each generated question into one of 5 categories. These are the percentage of questions that each model got correct for each category
You can use SQLCoder via the `transformers` library by downloading our model weights from the HuggingFace repo. We have added sample code for inference here. You can also use a demo on our website [here](https://defog.ai/sqlcoder).