-
Download data that are not available via API calls as instructed in data/data_preparation_guide.md.
-
Prepare the complete table data by create a master data file with data from data file downloaded and Yahoo! Finance.
python prepare_table_data.pyThe complete data could be found in file data/all_table_data.csv
-
Construct dataset by Pair the report with the corresponding tabular data and output as json file. The constructed data includes three subsets: train/test/validation. And process dataset for fine-tuning:
python construct_dataset.py
- Finetune model with DataTales:
python finetune_llm.py
- Run inference with DataTales test set:
python inference.py
- Reasoning evaluation:
python reasoning_evaluation.py - Style analysis:
python style_evaluation.py