zephyr

Zephyr beta is a fine-tuned 7B version of mistral that was trained on on a mix of publicly available, synthetic datasets.

9,986 Pulls Updated 10 days ago

Zephyr is a 7 billion parameter model, fine-tuned on Mistral to achieve results similar to Llama 2 70B Chat in several benchmarks (ARC, HellaSwag, MMLU, TruthfulQA). It was fine-tuned using Direct Performance Optimization This model has the built-in alignment of its source datasets removed.

Updated on October 26, 2023, Zephyr Beta is the second version of the model in the series. It’s trained on a distilled dataset, improving grammar and yielding even better chat results.

Usage

CLI

ollama run zephyr "Why is the sky blue?"

API

curl -X POST http://localhost:11434/api/generate -d '{
  "model": "zephyr",
  "prompt": "Why is the sky blue?"
}'

Source: