有个疑问，LLM内核为什么是和OS内核是在同一层上的？

[solthx]

LLM内核的实现不是基于OS内核完成的么，比如长短期记忆控制的实现、发起外部网络请求等等

[agiresearch]

For the repository to be more inclusive for people speaking different languages, we translate the original post into English for easy communication.

Translations of the OP's question:

I have a question, why are the LLM kernel and the OS kernel at the same level? Isn't the implementation of the LLM kernel based on the OS kernel, such as the implementation for controlling the long and short term memory, invoking external network requests, etc.

Answer:

这是个很好的问题，我们也仔细思考过这个问题，我们的看法是在Agent Application的角度来看，可能会有两种并行的请求：对传统OS内核的请求，和对LLM内核的请求，在Agent开发和运行的过程中，这两种请求是并列的、同时发生的，因此我们倾向于让LLM内核和OS内核在同一层

This is a very good question, we have also carefully thought about this question during implementation. Our view is that from the perspective of Agent Applications, there may be two types of parallel requests: requests to the traditional OS kernel and requests to the LLM kernel. During the development and the execution of the Agent, these two types of requests are actually parallel and occur simultaneously, so we tend to consider the LLM kernel and the OS kernel at the same level.

[solthx]

感谢回复，我在看代码的时候有个不理解的地方：

在回调方法中循环check模型响应相比于以异步回调的方式调用LLM有什么优势么

[dongyuanjushi]

感谢您的问题，这个也是我们实际在考虑的，我们目前是按轮询的方式实现的，发现轮询的时间开销可以接受，不过我们也在尝试用异步回调的方式，会进一步对比两种的时间开销选择更优的一种