Hi @ximinng,

Thanks for releasing the SVGX datasets. A few small questions about the datasets:

According to the README:

Available Datasets on Hugging Face:

• xingxm/SVGX-Core-250k: Core pretraining data (250k examples).
• xingxm/SVGX-SFT-1M: Supervised fine-tuning data (1M examples).

But I presume SVGX-Core-250k is the actual SFT dataset needed for fine-tuning, based on how it's used in the code? Could you clarify this discrepancy?

Also, could you explain what SVGX_SFT_GEN_51k, SVGX_SFT_GEN_basic, and SVGX_SFT_UN_25k in SVGX-SFT-1M are exactly? I've noticed that a lot of SVGs in SVGX-SFT-1M dataset appear to be broken.

e.g.