Please sign in to view datasets.
Pipeline and custom training datasets
Please sign in to view datasets.
Use a train-ready JSON or JSONL file as-is. No AI generation.
JSON or JSONL, 10–100,000 rows. Supported SFT shapes (fields kept as uploaded):
messages;
instruction (+ optional input) + output/response;
question/answer;
prompt/completion or response;
ShareGPT conversations;
text/label; single text;
input/output;
query/response;
human/assistant;
flat system/user/assistant.
Download sample