MT-Bench

A high-quality DPO derivative of LDJnr's Capybara, with chosen/rejected pairs synthesized and rated using Argilla's distilabel pipeline.

LDJnr's multi-turn reasoning dataset built with the Amplify-Instruct synthesis method - short but deep conversations on a single topic.

NVIDIA's permissively-licensed human-annotated preference dataset with 5-axis Likert ratings - engineered to train high-quality reward models.

A self-synthesis method (and family of datasets) that elicits high-quality instructions directly from an aligned LLM using only its chat template - no seed prompts required.

Berkeley NEST's seven-way ranked preference dataset built from GPT-4 rankings over responses from a diverse model pool, used to train Starling.

Teknium's million-row aggregation of high-quality GPT-4-style synthetic instructions that became the de facto open SFT baseline of 2023-2024.

The original community-scraped corpus of ChatGPT conversations that bootstrapped Vicuna and the entire open-instruction-tuning era.

Tsinghua / OpenBMB's large-scale multi-turn dialog dataset generated by two LLMs talking to each other across structured topic taxonomies.

OpenBMB's 64k-prompt preference dataset built with GPT-4 critiques across instruction-following, truthfulness, honesty, and helpfulness - the de facto open DPO baseline.

1M real user-chatbot conversations collected by Allen AI / UW via a free GPT proxy - a window into how real users actually prompt LLMs.

Microsoft's "Evol-Instruct" recipe - automatically rewriting simple instructions into harder, more diverse ones using an LLM evolver.

Introduces MT-Bench and validates LLM-as-a-Judge by showing GPT-4 judgments agree with humans at ~85%, the level of human-to-human agreement.

Introduces MT-Bench and validates LLM-as-a-Judge by showing GPT-4 judgments agree with humans at ~85%, the level of human-to-human agreement.