README.md

---
license: apache-2.0
language:
- arr
pretty_name: TrackMania Text
size_categories:
- 1M<n<10M
tags:
- driving
task_categories:
- text-classification
---

# tracktext

An experimental dataset that contains 128x64 greyscale images of TrackMania gameplay + keystrokes, designed for LLMs with 16k context or above.

Inspired by DOOM-Mistral-7b :)

## Dataset Details

Format:
```
{data}
[0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0],
...
{action}
[0, 0, 0, 0]
```

Greyscale Precision: 1 decimal
Capture rate: 6 frames per second

### Dataset Description

- **Curated by:** leafspark
- **Language(s) (NLP):** Numbers
- **License:** apache-2.0

## Uses

To train an LLM how to play TrackMania.

### Direct Use

Specialized LLM

### Out-of-Scope Use

Any regular LLM

## Dataset Structure

Text files

## Dataset Creation

Python script in repo

### Curation Rationale

Self driving using an LLM? For fun

### Source Data

TrackMania 2020 screengrabs

#### Personal and Sensitive Information

None

## Bias, Risks, and Limitations

The resolution is not very high; there may be suboptimal results

### Recommendations

Don't expect anything good