README.md

### Contents:
More thoroughly analysed and cleaned dataset and recently obtained. Extended the instruction prompt to hopefully get a slightly higher quality output. Data is synthetically generated with Gemini 2.0 Flash Lite, cost me about ~$1 to process 3k+ rows. v0.2 dataset currently only has 1 single type of dataset which is only directly collected from Rayman fist listings, I'm planning to mix it with other datasets i.e. chat dataset, non-Rayman fist listing dataset if v0.2 models shows improved performance than v0 models.

### Modifications:
- Nerfed currency extrapolation, because current v0 models has very low success rates of accurately predicting item currency based on prices (70% accurate), and are biased towards extrapolating to "DL" currency most of the time. So instead im focusing solely on info extraction