Skip to content

Commit

Permalink
add how-openai-o1-strawberry been trained
Browse files Browse the repository at this point in the history
  • Loading branch information
wjmcater committed Sep 15, 2024
1 parent f0d14d1 commit cfc5cdd
Show file tree
Hide file tree
Showing 4 changed files with 32 additions and 0 deletions.
6 changes: 6 additions & 0 deletions docs/01-llm-intro/06-function-call.mdx
Original file line number Diff line number Diff line change
@@ -0,0 +1,6 @@
---
title: "工具调用(Function Call)"
---


sdfsdf
17 changes: 17 additions & 0 deletions docs/02-prompt-engineering/02-chain-of-thought/03-concise-cot.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,17 @@
---
title: "Concise Thought"
---


CoT 的长度通常是会影响整体时延,有的项目当中会直接把 CoT 的策略给删除,可是这个通常会影响模型的效果.

此时你可以有两种策略:
1. 增加你的训练数据
2. 缩短你的 CoT 长度进而提升模型的效果.

第一种方法就非常的简单粗暴,不过很可能会破坏你这个模型的通用 chat 能力,除非你的目的就是要构建一个垂类的不具备很强通用 chat 能力的模型。


## 参考文章

* [[1] Concise Thoughts: Impact of Output Length on LLM Reasoning and Cost](h是)
9 changes: 9 additions & 0 deletions docs/03-agent-application/how-does-openai-o1-train.mdx
Original file line number Diff line number Diff line change
@@ -0,0 +1,9 @@
---
title: "OpenAI 的O1 模型是如何训练的"
---

或许训练的模式是这个样子:

![](./imgs/strawberry-training-mode.gif)

> 此图片来源于外网一位大佬
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.

0 comments on commit cfc5cdd

Please sign in to comment.