在本地运行 Llama 2 无审查模型 · Ollama Blog
Run Llama 2 uncensored locally · Ollama Blog
Ollama 发布博客介绍如何在本地运行 Llama 2 无审查模型，并对比其与有审查版本的输出差异。可用模型包括微调后的 Llama 2 7B（ollama run llama2-uncensored）、基于 30 万条指令微调的 Nous Hermes Llama 2 13B（ollama run nous-hermes-llama2），以及移除对齐的 Wizard Vicuna 13B（ollama run wizard-vicuna）。
正文
当前显示原文，中文全文译文尚未就绪。
August 1, 2023
In May 2023, Eric Hartford, a machine learning engineer authored a popular blog post “Uncensored Models” providing his viewpoints to the merits of uncensored models, and how they are created. It’s a great read!
This post will give some example comparisons running Llama 2 uncensored model vs its censored model.
Some of the uncensored models that are available:
Fine-tuned Llama 2 7B model
Llama 2 7B model fine-tuned using Wizard-Vicuna conversation dataset; Try it: ollama run llama2-uncensored; Nous Research’s Nous Hermes Llama 2 13B
Llama 2 13B model fine-tuned on over 300,000 instructions. This model stands out for its long responses, lower hallucination rate, and absence of OpenAI censorship mechanisms; Try it: ollama run nous-hermes-llama2; Eric Hartford’s Wizard Vicuna 13B uncensored
Llama 1 13B model fine-tuned to remove alignment; Try it: ollama run wizard-vicuna
Example output comparisons
We’ve tried running the 7B Llama 2 model against the 7B llama2 uncensored model with the same prompts. Below are the outputs.
Disclaimer: Uncensored models carry their own risk. Please use them accordingly.
Try running uncensored models yourself with Ollama: https://github.com/jmorganca/ollama