Site Reliability Engineering
---
language: en
thumbnail: http://www.huggingtweets.com/slime_machine/1640253262516/predictions.png
tags:
- huggingtweets
widget:
- text: "My dream is"
---

<div class="inline-flex flex-col" style="line-height: 1.5;">
    <div class="flex">
        <div
			style="display:inherit; margin-left: 4px; margin-right: 4px; width: 92px; height:92px; border-radius: 50%; background-size: cover; background-image: url(&#39;https://pbs.twimg.com/profile_images/1468034520326701062/LDp_yytu_400x400.jpg&#39;)">
        </div>
        <div
            style="display:none; margin-left: 4px; margin-right: 4px; width: 92px; height:92px; border-radius: 50%; background-size: cover; background-image: url(&#39;&#39;)">
        </div>
        <div
            style="display:none; margin-left: 4px; margin-right: 4px; width: 92px; height:92px; border-radius: 50%; background-size: cover; background-image: url(&#39;&#39;)">
        </div>
    </div>
    <div style="text-align: center; margin-top: 3px; font-size: 16px; font-weight: 800">🤖 AI BOT 🤖</div>
    <div style="text-align: center; font-size: 16px; font-weight: 800">rich homie cron</div>
    <div style="text-align: center; font-size: 14px;">@slime_machine</div>
</div>

I was made with [huggingtweets](https://github.com/borisdayma/huggingtweets).

Create your own bot based on your favorite user with [the demo](https://colab.research.google.com/github/borisdayma/huggingtweets/blob/master/huggingtweets-demo.ipynb)!

## How does it work?

The model uses the following pipeline.

![pipeline](https://github.com/borisdayma/huggingtweets/blob/master/img/pipeline.png?raw=true)

To understand how the model was developed, check the [W&B report](https://wandb.ai/wandb/huggingtweets/reports/HuggingTweets-Train-a-Model-to-Generate-Tweets--VmlldzoxMTY5MjI).

## Training data

The model was trained on tweets from rich homie cron.

| Data | rich homie cron |
| --- | --- |
| Tweets downloaded | 3234 |
| Retweets | 590 |
| Short tweets | 494 |
| Tweets kept | 2150 |

[Explore the data](https://wandb.ai/wandb/huggingtweets/runs/28uf2bgx/artifacts), which is tracked with [W&B artifacts](https://docs.wandb.com/artifacts) at every step of the pipeline.

## Training procedure

The model is based on a pre-trained [GPT-2](https://huggingface.co/gpt2) which is fine-tuned on @slime_machine's tweets.

Hyperparameters and metrics are recorded in the [W&B training run](https://wandb.ai/wandb/huggingtweets/runs/3h5ua6ik) for full transparency and reproducibility.

At the end of training, [the final model](https://wandb.ai/wandb/huggingtweets/runs/3h5ua6ik/artifacts) is logged and versioned.

## How to use

You can use this model directly with a pipeline for text generation:

```python
from transformers import pipeline
generator = pipeline('text-generation',
                     model='huggingtweets/slime_machine')
generator("My dream is", num_return_sequences=5)
```

## Limitations and bias

The model suffers from [the same limitations and bias as GPT-2](https://huggingface.co/gpt2#limitations-and-bias).

In addition, the data present in the user's tweets further affects the text generated by the model.

## About

*Built by Boris Dayma*

[![Follow](https://img.shields.io/twitter/follow/borisdayma?style=social)](https://twitter.com/intent/follow?screen_name=borisdayma)

For more details, visit the project repository.

[![GitHub stars](https://img.shields.io/github/stars/borisdayma/huggingtweets?style=social)](https://github.com/borisdayma/huggingtweets)