by Teknium
OpenHermes 2.5 is a dataset of about 1M mostly synthetic instruction and chat samples compiled by Teknium. It is the data behind the OpenHermes 2.5 and Nous Hermes 2 models.
DatasetFree
Step 1: Open the dataset page on Hugging Face and read the card.
Step 2: Check the listed source datasets and their terms.
Step 3: Download the dataset and load the conversations field of each entry.
Step 4: Map the from and value fields to your training format.
Step 5: Optionally use the source and category fields to filter or balance the data.
A compilation of about 1M mostly synthetic instruction and chat samples. It underpins the OpenHermes 2.5 and Nous Hermes 2 models.
ShareGPT structure. Each entry has a conversations list, and each turn has a from field for the role and a value field for the text.
Many open source datasets and custom synthetic ones. The card lists each source with a link.
Yes. It is hosted on Lilac's Hugging Face space for exploration, search and clustering.
HH-RLHFHH-RLHF is Anthropic's dataset of human preference comparisons on helpfulness and harmlessness, plus red teaming transcripts. It is meant for research on making models less harmful.