31 lines
1.3 KiB
Markdown
31 lines
1.3 KiB
Markdown
<h1 align="center">miniLLMFlow</h1>
|
||
|
||
|
||
|
||
Minimalist LLM framework in [100 lines](minillmflow/__init__.py). Express popular paradigms like agents, task decomposition, RAG, and more.
|
||
|
||
- Install via ```pip install minillmflow```. Or just copy the [source codes](minillmflow/__init__.py) (it's only 100 lines)
|
||
- We **strongly recommened** using LLMs (e.g., Claude, ChatGPT, Cursor) to develop LLM applications with [this prompt](minillmflow/docs/prompt)
|
||
|
||
## Why miniLLMFlow?
|
||
|
||
The future of programming will be heavily LLM-assited, and LLMs:
|
||
|
||
1. **😀 Shine at Feature Implementation**:
|
||
With proper docs, LLMs can handle APIs, tools, chunking, prompt wrapping, etc.
|
||
These are hard to maintain and optimize, so they don’t belong in the framework.
|
||
|
||
2. **☹️ Struggle with Paradigm Design**:
|
||
Paradigms like MapReduce, task decomposition, and agents are powerful, even for LLMs.
|
||
However, designing these elegantly remains challenging for LLMs.
|
||
|
||
To enable LLMs to develop LLM applications, a framework should
|
||
(1) remove redunant feature implementations but
|
||
(2) keep core paradigms to build on.
|
||
It turns out that such a framework only needs 100 lines of code.
|
||
|
||
<div align="center">
|
||
<img src="./docs/minillmflow.jpg" width="400"/>
|
||
</div>
|
||
|