AI "tuned" and condensed dataset.
The org dataset was "polaris" which is a general use dataset generated by GPT 5.
This dataset was run thru Qwen 3.6 35B-A3B (Q6) via LMStudio (local server) and optimized.
The prompts are exactly the same, just the answers have be updated/modified/optimized.
Each answer was evaluted by Qwen model, then optimized/adjusted and/or corrected.
The dataset SIZE has shrunk by 25% appox (5.3 MBs vs 3.5 MBs).
This is a test case in part to eval compression of answers… See the full description on the dataset page:
https://huggingface.co/datasets/ansulev/general-training-el-supremo.
