https://github.com/jovanveljanoski/openai-gpt-logprobs-examples

Examples of using the logprobs from a GPT response demoed as a simple solara app
https://github.com/jovanveljanoski/openai-gpt-logprobs-examples

Last synced: 4 days ago
JSON representation

Examples of using the logprobs from a GPT response demoed as a simple solara app

• Host: GitHub
• URL: https://github.com/jovanveljanoski/openai-gpt-logprobs-examples
• Owner: JovanVeljanoski
• License: mit
• Created: 2024-08-23T13:36:43.000Z (4 months ago)
• Default Branch: master
• Last Pushed: 2024-08-24T15:52:34.000Z (4 months ago)
• Last Synced: 2024-10-31T20:43:47.131Z (about 2 months ago)
• Language: Python
• Homepage:
• Size: 20.5 KB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
---

title: openai-gpt-logprobs-examples

emoji: 🚀

colorFrom: red

colorTo: green

sdk: docker

pinned: false

license: mit

app_port: 7860

---

### Simple examples of using `logprobs` from outputs of OpenAI GPT models

This repo contains some simple examples of how to use, display and use `logprobs`

as part of the output of GPT models from OpenAI.

It runs as a simple [solara](https://solara.dev/) web-application with 3 tabs:

1. Get the `logprobs` per word, from a single reponse from a LLM model, coloured by their probability. This way you can see the possible variations in the answer you are getting.

2. A binary classification example. The reply is a `True` or `False`. The probability is extracted from the `logprob` of the `True/False` token only. Can be used to "tune" your classifier based on this probability, much like you can do with a classical ML classifier.

3. A simple RAG example. It combines the elements of both examples above.

### Links

- [GitHub repo](https://github.com/JovanVeljanoski/openai-gpt-logprobs-examples)

- [Huggingface space](https://huggingface.co/spaces/Jovan31/openai-gpt-logprobs-examples)

### To install / use locally

Clone this repo locally:

```bash

git clone [email protected]:JovanVeljanoski/openai-gpt-logprobs-examples.git

```

Install the dependencies, preferrably in a new (virtual) environment:

```bash

pip install -r requirements.txt

```

Launch the app

```bash

solara run app.py

```