https://github.com/jdinkla/ai-playground
Examples for OpenAI and langchain using Python
https://github.com/jdinkla/ai-playground
dialogue-generation langchain langchain-python openai openai-api python python3
Last synced: 4 months ago
JSON representation
Examples for OpenAI and langchain using Python
- Host: GitHub
- URL: https://github.com/jdinkla/ai-playground
- Owner: jdinkla
- License: mit
- Created: 2023-04-07T16:19:23.000Z (about 2 years ago)
- Default Branch: main
- Last Pushed: 2024-06-25T12:45:42.000Z (10 months ago)
- Last Synced: 2024-08-13T07:08:01.460Z (8 months ago)
- Topics: dialogue-generation, langchain, langchain-python, openai, openai-api, python, python3
- Language: Python
- Homepage:
- Size: 79.1 KB
- Stars: 1
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- jimsghstars - jdinkla/ai-playground - Examples for OpenAI and langchain using Python (Python)
README
# ai-playground
This project contains my examples exploring the APIs of [OpenAI](https://platform.openai.com/playground)
and [langchain](https://python.langchain.com/docs/get_started/introduction).
The `src` folder contains example scripts.
Some of these scripts need command line arguments. You can call the program with the `-h` flag to get an explanation. For example
$ python src/openai_dialogue.py -h
usage: openai_dialogue.py [-h] filename turns [{gpt-3.5-turbo-1106,gpt-4-32k-0613,gpt-4-1106-preview}]
positional arguments:
filename the name of a json file
turns the number of turns
{gpt-3.5-turbo-1106,gpt-4-32k-0613,gpt-4-1106-preview}
the OpenAI model
options:
-h, --help show this help message and exit
There are the following examples
| File | Description |
| ------------- | ------------- |
| langchain_chain.py | Example for chain |
| langchain_conversation_chain.py | Example for a conversation |
| langchain_llm.py | Simple example |
| langchain_prompt_text2image.py | Example for prompt |
| langchain_rag.py | Example for chain with retrieval |
| langchain_rag_extended.py | Example for querying text with RAG |
| openai_chat.py | Example for chat |
| openai_completion.py | Example for completion |
| openai_detective.py | Solve a mystery, test for GTP4.5 Turbo |
| openai_dialogue_generic.py | Create a dialogue, see below |
| openai_image.py | Create an image from a prompt |
| openai_speak.py | Create audio from a prompt |
## Creating dialogues
A more advanced example is the creation of dialogues of two or more agents.
The scene of a dialog can be specified in a `yaml` or `json` file. See the `examples` folder and the next section.
The following example is specification of a scene.
```yaml
description: >
This is the balcony scene similiar to 'Romeo and Juliet' by William Shakespeare but in 2023.
Do not say more than 4 sentences.
persons:
- name: Romeo
prompt: You are Romeo but more like a gangsta rapper, you speak short and precise. You want a date with Juliet.
voice: onyx
- name: Juliet
prompt: You are Julie but more like a Curt Cobain and you do not like to talk much.
voice: nova
```
And this is an example session
$ python src/openai_dialogue.py examples/modern_romeo_and_juliet.yaml 2 "gpt-4-1106-preview"
----------------------------------------------------
[Romeo] Yo, Jules, peepin' ya still, queen of my world – fly high in them skies.
Light it up, my night’s delight; how 'bout we roll, just you and I?
Words ain’t enough, babe, gotta show; let's bounce, hit the town, it's our time.
Drop a pin, I’ll scoop you fast; together, we'll set this night on fire, make it sublime.
----------------------------------------------------
[Juliet] Yeah, it's me. What's up?
----------------------------------------------------
Yo, Jules, it's me, 'Romeo';
I'm down here dreamin' 'bout us, solo.
Holla at your window, let's not wait,
Slide down, girl, it's fate – don't be late.
----------------------------------------------------
[Juliet] Not much to say; let's flee the talk.
Beneath your pane, I'll meet you—walk?
This night's ours, no need for light.
Let's vanish, just feels right.
## Generating a dialogue descriptionv
With `src/openai_dialogue_generator.py` you can generate dialogue descriptions and modify them afterwards if needed.
$ python src/openai_dialogue_generator.py "Ernie and Bert discuss if they should eat cake or cookies"
---------------------------------------------
description: >
Ernie and Bert are having a discussion in their living room about what they should have for dessert - cake or cookies.
They are sitting on their couch, surrounded by their favorite toys and books.
persons:
- name: Ernie
prompt: You are Ernie, and you love the idea of having cake for dessert. Use excitement and exclamation in your speech to convince Bert.
voice: onyx
- name: Bert
prompt: You are Bert, and you prefer cookies over cake. Use a calm and reasoned tone to persuade Ernie that cookies are the better choice.
voice: nova
## Installation
Install with
$ make init
Set the OPENAI-API key.
$ export OPENAI_API_KEY=""
$ source venv/bin/activate
## Issues
### docarray needs old version of pydantic
The examples for RAG use `docarray` and this needs an old version of pydantic (at the time of writing).
$ pip install pydantic==1.10.9
$ python src/langchain_rag.py
Install the new version again with.
$ pip install -U pydantic
See also the [langchain wiki](https://python.langchain.com/docs/guides/pydantic_compatibility).