---
license: mit
tags:
  - prompt-routing
  - llm-routing
  - complexity-classification
  - text-classification
  - prompt-difficulty
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train.csv
---

# Skala-Dataset

The training corpus behind [Skala](https://huggingface.co/TakamaruAI/Skala), a prompt-difficulty classifier that labels chat, agent, and tooling prompts into three difficulty tiers to help routers select an appropriate model for each request.

## Data

| column | type | description |
|---|---|---|
| `text` | string | the prompt |
| `label` | string | `simple`, `medium` or `complex` |
| `source` | string | where the row came from |

| source | rows | licence |
|---|---:|---|
| `public_dataset_master` | 1,348 | MIT |
| `public_dataset_real` | 350 | MIT |
| `generated` | 5,242 | MIT |
| **total** | **6,940** | |

- **`public_dataset_master`** -
  [Nasim435/Multi-label-Prompt-Dataset](https://huggingface.co/datasets/Nasim435/Multi-label-Prompt-Dataset)
  (MIT), the `master_dataset.csv` file. Its complexity tags (`easy`/`moderate`/`hard`)
  map to `simple`/`medium`/`complex`. Filtered to rows whose multi-label set contains
  exactly one complexity tier, de-duplicated, with the evaluation rows excluded.
- **`public_dataset_real`** - The `real.csv` split of the same dataset, short developer
  and sysadmin questions.
- **`generated`** - Generated and labelled by two open-weight models, `google/gemma-4-26B-A4B` and `Qwen/Qwen3.8-27B`.

## Licence

Skala-Dataset is licensed under the [MIT License](LICENSE).