An open API service indexing awesome lists of open source software.

https://github.com/alvarobartt/opentrain

🚂 Fine-tune OpenAI models for text classification, question answering, and more
https://github.com/alvarobartt/opentrain

fine-tune openai openai-datasets openai-python

Last synced: 3 months ago
JSON representation

🚂 Fine-tune OpenAI models for text classification, question answering, and more

Awesome Lists containing this project

README

          


opentrain



🚂 Fine-tune OpenAI models for text classification, question answering, and more


---

`opentrain` is a simple Python package to fine-tune OpenAI models for task-specific purposes such as text classification, token classification, or question answering.

More information about OpenAI Fine-tuning at https://platform.openai.com/docs/guides/fine-tuning.

## 💻 Usage

### 📦 Data management

```python
import openai
from opentrain import Dataset

openai.api_key = ""

dataset = Dataset.from_file("data.jsonl")
dataset.info
dataset.download(output_path="downloaded-data.jsonl")
```

### 🦾 Fine-tune

```python
import openai
from opentrain import Train

openai.api_key = ""

trainer = Train(model="ada")
trainer.train(
[
{
"prompt": "I love to play soccer ->",
"completion": " soccer",
},
{
"prompt": "I love to play basketball ->",
"completion": " basketball",
},
],
)
```

### 🤖 Predict

```python
import openai
from opentrain import Inference

openai.api_key = ""

predict = Inference(model="ada:ft-personal-2021-03-01-00-00-01")
predict.predict("I love to play ->")
```

## ⚠️ Warning

Fine-tuning OpenAI models via their API may take too long, so please be patient. Also, bear in mind
that in some cases you just won't need to fine-tune an OpenAI model for your task.

To keep track of all the models you fine-tuned, you should visit https://platform.openai.com/account/usage,
and then in the "Daily usage breakdown (UTC)" you'll need to select the date where you triggered the
fine-tuning and click on "Fine-tune training" to see all the fine-tune training requests that you sent.

Besides that, in the OpenAI Playground at https://platform.openai.com/playground, you'll see a dropdown
menu for all the available models, both the default ones and the ones you fine-tuned. Usually, in the
following format `:ft-personal-`, e.g. `ada:ft-personal-2021-03-01-00-00-01`.