Kishor V2 is a diverse and compact dataset curated for training small to medium-sized language models. It includes 53,000 structured prompt-response pairs across multiple domains to simulate human-like dialogue, reasoning, and general intelligence.
KishorV2_dataset.jsonl: Main dataset in JSON Lines format.
Each line is a JSON object with:
{
"type": "qa" | "dialogue" | "quote" | "fact" | "reasoning" |… See the full description on the dataset page:
https://huggingface.co/datasets/GODELEV/Kishor_V2_53K_LLM_Prompt-Response_Pairs.
