This repository contains the datasets used for training (TrainData) and testing (test.jsonl) the Turkish SQL dataset, as well as modified codes.
Training was conducted using the SuperAdapters repository.
The InferenceBatch file requires the SuperAdapters library to run. The InferenceBatch file is a modified version of the inference code from the SuperAdapters repository to obtain batch results.
The fine-tuned models and their links are as follows:
| Model | Link |
|---|---|
| Llama 2 SQL | emirozturk/Llama-2-7b-chat-hf-SQL |
| Llama 2 Turkish | emirozturk/Llama-2-7b-chat-hf-Turkish |
| Llama 2 Turkish SQL | emirozturk/Llama-2-7b-chat-hf-Turkish-SQL |
| Llama 3 SQL | emirozturk/llama3-Instruct-SQL |
| Llama 3 Turkish | emirozturk/llama3-Instruct-Turkish |
| Llama 3 Turkish SQL | emirozturk/llama3-Instruct-Turkish-SQL |
| Phi 3 SQL | emirozturk/Phi-3-mini-4k-instruct-SQL |
| Phi 3 Turkish | emirozturk/Phi-3-mini-4k-instruct-Turkish |
| Phi 3 Turkish SQL | emirozturk/Phi-3-mini-4k-instruct-Turkish-SQL |
InferenceBatchfile is a modified version of the inference code from the SuperAdapters repository to obtain batch results.calculateExAndLfAcc.pycalculates Execution Accuracy and LF accuracy values using the Tur2SQL database (data.db) and wikiSQL Evaluation codes.