ChatML-formatted instruction-tuning dataset for SEC 10-K financial document question answering. Used to train finlora-sft-v2-phi35 via QLoRA supervised fine-tuning on Phi-3.5-mini-instruct.

	Dataset summary

Split Examples Size

train 31,320 ~99 MB

validation 3,479 ~11 MB

Total: 34,799 examples

Sources

EDGAR (28,548 examples) — 125 SEC 10-K annual reports across 25 S&P 500 companies, 5 sectors (Technology, Finance… See the full description on the dataset page: https://huggingface.co/datasets/AayushP418/finlora-sft-dataset.