| --- |
| license: cc-by-4.0 |
| task_categories: |
| - text-generation |
| language: |
| - en |
| pretty_name: Chess Instruct |
| size_categories: |
| - 10K<n<100K |
| --- |
| |
| ## ChessInstruct |
|
|
| The ChessInstruct Dataset serves as the foundation for training and fine-tuning Language Models (LLMs) specifically in the realm of chess instruction. |
| Derived from the [laion/strategic_game_chess](https://huggingface.co/datasets/laion/strategic_game_chess) dataset, this meticulously curated dataset encompasses a wide array of annotated instructional chess content. |
|
|
| Features of the ChessInstruct Dataset: |
|
|
| * **Rich and Diverse Content**: Curated with a broad spectrum of instructional resources including annotated games, strategic analyses (incoming) and positional evaluations, the dataset facilitates comprehensive learning and modeling. |
| * **Customizable Training Resource**: The ChessInstruct Dataset allows for the tailored fine-tuning of any Language Model, enabling researchers and practitioners to adapt and optimize LLMs for chess-specific instructional contexts. |
| * **Annotated Instructional Insights**: Detailed annotations and instructional cues within the dataset provide valuable guidance for language model training, emphasizing strategic moves, tactics, and decision-making processes. |
|
|
| ## Usage |
|
|
| The ChessInstruct dataset comprises four primary columns: |
| * `task`: This column contains instruct prompts related to various chess scenarios, such as predicting the winner given a set of chess moves or identifying the last move in a sequence. |
| * `input`: The input column provides supplementary information, usually a series of chess moves, to support the instruct prompt. These inputs are presented as JSON-serialized strings. |
| * `expected_output`: This column presents the anticipated or expected output corresponding to the instruct task. The expected outputs are also serialized as JSON strings. |
| * `KIND`: The KIND column categorizes the type of instruct prompt, delineating the nature of the task, whether it involves identifying winning scenarios, predicting subsequent moves, or performing other chess-related analyses. |
|
|
| ### Distribution |
| | Task | Number of samples training set | Number of samples test set | Distribution | |
| |------|--------------------------------|----------------------------|--------------| |
| | Finding last movement | 13500 | 1500 |15% | |
| | Finding game's score | 18000 | 2000 | 20% | |
| | Finding missing movements | 13500 | 1500 | 15% | |
| | Finding the best possible move to do | 18000 | 2000 | 20% | |
| | Finding who is advantaged in the game | 18000 | 2000 | 20% | |
| | Sorting FENs from earliest to oldest in the game | 9000 | 1000 | 10% | |
|
|
|
|
| ## Reproduction |
|
|
| All the necessary code to reproduce this dataset is available here: [Thytu/StockLLM](https://github.com/Thytu/StockLLM) |
|
|
| ## Citation |
|
|
| This dataset is based on [laion/strategic_game_chess](https://huggingface.co/datasets/laion/strategic_game_chess?row=0) which I thank dearly for the data |
