search

tingxueronghua / ChartLlama-code

MIT License
178 stars 17 forks source link

Hyperparameters for training your model on Chart2Text train dataset #13

Closed y-rok closed 7 months ago

y-rok commented 7 months ago

Hello.

In your research paper, you mentioned that you trained ChartLlama on the train split when evaluating using BLEU-4. Can you tell me what hyper-parameters you used to train the model for Chart2text? (e.g. epochs, batch size...)

Thanks in advance :)

tingxueronghua commented 7 months ago

I use the same training hyperparameters except the epochs as 10 for pew and 3 for statistas, as I remember. In fact I do not have much time for choosing the hyperparameters, and I think you can try more to get higher performances.