Closed cpcdoy closed 1 year ago
Hi @cpcdoy, thanks for your interests in our work.
We do not provide the generated NL datasets, but only introduce how to make them using Vega-Lite specs.
Although you can find sample results for 48 charts that are selected via stratified sampling: https://github.com/hyungkwonko/chart-llm/blob/main/exp/gold/result/gold.csv
You can generate the L1/L2 captions, utterances, questions using Vega-Lite and our prompting technique: https://github.com/hyungkwonko/chart-llm#example-nl-dataset-generation
The main reason we only test it on 48 samples is, you know, it costs a lot.
Let me know if you have any further question.
Hi,
Awesome work on this, I looked quite a bit in the repo and couldn't find the generated NL prompts, questions, etc for the 1981 charts that you mention in the paper. I only found the description field sometimes filled in the chart and some example runs, but that's it.
Could you direct me to the final dataset?
Thanks!