BrewInteractive/alpaca-tr

---
license: apache-2.0
task_categories:
- text-generation
language:
- tr
size_categories:
- 10K<n<100K
---

## Dataset Description

**Dataset Name:** Turkish Alpaca Dataset

**Languages:** Turkish

**Data Source:**
- Original dataset created from Alpaca GPT-4.
- Translated and localized using Gemini Flash.
- Most generated texts from the Alpaca dataset were modified, retaining only the instructions.
- Cleaned to remove non-Turkish texts.
- Irrelevant contexts were also cleaned.

### Dataset Summary

This dataset includes a series of text instructions originally sourced from the Alpaca GPT-4 dataset. The instructions were translated and localized into Turkish using Gemini Flash, with significant modifications made to the generated texts. The dataset emphasizes contextually relevant and locale-specific information, suitable for fine-tuning small language models to be aware of and responsive to Turkish language and cultural contexts.

## Motivation

### Why We Created This Dataset

Our goal is to develop, fine-tune (and debug?) small language models that are both contextually and locale-aware.

1. **Localized Training Data:** Essential for models intended for Turkish language applications.
2. **Context-Aware Examples:** Ensures that the models are capable of understanding and generating contextually relevant responses.
3. **Evaluation Resource:** Aids in testing and validating the performance of smaller models in terms of handling Turkish instructions effectively.

### Intended Uses

This dataset is primarily intended for fine-tuning and testing small-scale language models to:

- Improve Turkish language understanding.
- Generate Turkish text that is contextually accurate and culturally relevant.
- Benchmark performance in localized NLP tasks.

## Data Collection and Processing

This dataset will be updated/cleaned regularly to ensure ongoing relevance and improved coverage of Turkish language instructions.

### Data Sources

- Alpaca GPT-4: Original source for instructions.
- Gemini Flash: Used for translation and localization processes.

**Dataset Curator:**
Brew Interactive/AI Guild
https://brewww.com