展示HN：Omni-NLI – 一个多接口的自然语言推理服务器
大家好，
我开发了一个名为 Omni-NLI 的开源工具，用于自然语言推理。它可以使用不同的模型来检查一段文本（称为前提）是否支持另一段文本（称为假设）。这种工具的主要应用是进行软事实核查和文本之间的一致性检查，例如句子之间的关系。
目前，Omni-NLI 具备以下功能：
- 可以通过 `pip install omni-nli[huggingface]` 安装为 Python 包。
- 可以在您的个人计算机上使用，因此您的数据保持本地和私密。
- 具有 MCP 接口（用于代理）和 REST API，适合作为微服务的常规使用。
- 支持使用来自不同来源的模型（Ollama、OpenRouter 和 HuggingFace）。
- 可以用来检查模型是否存在自我矛盾的情况。
- 支持展示推理过程，让您了解它为何认为某个主张是错误的。
如果您有兴趣了解更多信息，以下是相关链接：
项目的 GitHub 仓库：[https://github.com/CogitatorTech/omni-nli](https://github.com/CogitatorTech/omni-nli)
项目的文档：[https://cogitatortech.github.io/omni-nli/](https://cogitatortech.github.io/omni-nli/)
查看原文
Hi everyone,<p>I've made an open-source tool (called Omni-NLI) for natural language inference. It can use different models to check if
a piece of text (called a premise) supports another piece of text (a hypothesis). The main application of a tool like
this is for soft fact-checking and consistency checking between pieces of texts like sentences.<p>Currently, Omni-NLI has the following features:<p><pre><code> - Can be installed as a Python package with `pip install omni-nli[huggingface]`.
- Can be used on your own computer, so your data stays local and private.
- Has an MCP interface (for agents) and a REST API for conventional use as a microservice.
- Supports using models from different sources (Ollama, OpenRouter, and HuggingFace).
- Can be used to check if it seems that a model is contradicting itself.
- Supports showing the reasoning so you can see why it thinks a claim is wrong.
</code></pre>
In any case, if you are interested in knowing more, there is more information in the links below:<p>Project's GitHub repo: <a href="https://github.com/CogitatorTech/omni-nli" rel="nofollow">https://github.com/CogitatorTech/omni-nli</a><p>Project's documentation: <a href="https://cogitatortech.github.io/omni-nli/" rel="nofollow">https://cogitatortech.github.io/omni-nli/</a>