| --- |
| license: apache-2.0 |
| language: |
| - en |
| task_categories: |
| - text-generation |
| tags: |
| - code |
| - agentic |
| - chain-of-thought |
| - sft |
| - synthetic |
| - distillation |
| pretty_name: '-' |
| size_categories: |
| - 1K<n<10K |
| --- |
| |
| # 🤖 Agentic Coding CoT Dataset v1.1 |
|
|
| A high-quality supervised fine-tuning (SFT) dataset for training agentic coding assistants with Chain-of-Thought reasoning capabilities. |
|
|
| ## 📋 Dataset Description |
|
|
| This dataset was created by processing and distilling ~20GB of GitHub crawl data using **Minimax-M2 & MiniMax M2.1** to generate structured, reasoning-rich coding examples. Each sample demonstrates systematic problem-solving with explicit tool usage patterns. |
|
|
| ## 🏗️ Assistant Data Structure |
|
|
| Assistant row contains a JSON object with the following fields: |
|
|
| | Field | Description | |
| |-------|-------------| |
| | `task` | Clear description of the coding task | |
| | `context` | Background information, schemas, or specifications | |
| | `plan` | Strategic breakdown (3-8 steps) | |
| | `cot` | Chain-of-Thought reasoning explaining the approach | |
| | `actions` | Sequence of tool calls (editor, bash, python, browser) | |
| | `final_answer` | Summary of implementation and key features | |
|
|
| ## 🛠️ Generation Pipeline |
| ``` |
| GitHub Crawl (~20GB) → Filtering & Cleaning → Distillation (Minimax-M2) → SFT Formatting → Quality Filtering |
| ``` |
|
|
| ## 💡 Intended Use |
|
|
| - Fine-tuning LLMs for agentic coding tasks |
| - Training models with structured reasoning capabilities |
| - Developing AI coding assistants with tool-use abilities |
|
|
| ## 📊 Example Tools in Dataset |
|
|
| - **editor**: File creation and modification |
| - **bash**: Shell command execution |
| - **python**: Code validation and testing |
| - **browser**: Documentation lookup |
|
|
|
|
| ## ⚠️ Limitations |
|
|
| - Synthetically generated via distillation |
| - English-only content |
| - May contain model-inherited biases |
|
|
| ## 📧 Contact |
|
|
| For questions or feedback, reach out via [X](https://x.com/AlicanKiraz0) or open an issue. |
