Tw-LeetCode

---

Overview

This dataset contains the solutions to the problems on LeetCode.

• Update Frequency: At least one problem every day.
• Language: TypeScript
• Model: Solutions and explanations are generated using Anthropic's Claude Sonnet 4.6 (Old solution use OpenAI's GPT-5 and GPT-5-Thinking models) with human cleanup.
• Question Code Source: TypeScript starter code is extracted from the whiskwhite/leetcode-complete dataset. Missing problems not covered by this dataset are manually collected to ensure completeness.

---

Solution Optimization Process

The approach to solving and optimizing LeetCode problems follows this process with optimization_prompt.md prompt:

1. Create a basic solution that correctly solves the problem
2. Leverage Claude 4.6 Sonnet model to iteratively optimize the solution using a specialized prompt

The optimization flow works as follows:

flowchart TD
    A[Create Basic Solution] --> B[Submit to LeetCode]
    B --> C[Record Execution Time/Performance]
    C --> D[Send to o4-mini-high with Optimization Prompt]
    D --> E[Get Optimized Solution]
    E --> F[Submit to LeetCode]
    F --> G{Is Performance Better?}
    G -->|Yes| H[Set as Current Best Solution]
    H --> D1[Start New Chat Session]
    D1 --> D
    G -->|No| I{Execution Error?}
    I -->|Yes| J[Get Error Feedback]
    J --> E
    I -->|No| K[Provide Performance Feedback]
    K --> E
    G -->|Same Performance & Beats 100%| L[Stop Optimization]
    L --> M[Generate Solution Explanation]
    M --> N[Refine Solution with GPT-4.5/4.1]
    N --> O[Manual Human Cleanup]
    
    style A fill:#d0f0c0
    style L fill:#f0c0c0
    style H fill:#c0d0f0
    style D1 fill:#c0f0f0
    style M fill:#f0e0c0
    style N fill:#e0c0f0
    style O fill:#c0e0f0

Loading

---

Optimization Process

The optimization process continues iteratively with these important rules:

• When a better solution is found, we always start a new chat session with the improved solution as the new baseline
• Error fixes and performance feedback continue in the same chat session
• We only stop iterations when the solution reaches 100% runtime efficiency AND new optimization attempts yield the same performance (bottleneck reached)
• The process aims to push solutions to their theoretical limits through continuous refinement

---

Solution Explanation and Cleanup

After optimization is complete, the following additional steps are performed:

1. Generate Solution Explanation: Use the solution prompt in solution_prompt.md with the Claude 4.6 Sonnet model to create detailed solution steps and explanations

2. Solution Refinement: Leverage Claude 4.6 Sonnet to refine the generated solution explanations for clarity and completeness

3. Manual Review and Cleanup: Perform a human review of the solution markdown to ensure:

   • Technical correctness of the solution
   • Proper formatting of the output
   • Clarity of explanations
   • Consistency with the repository standards

---

Citation

If you find this dataset useful, please consider citing it in your work:

@misc{twleetcode2025,
  title        = {tw-leetcode: Traditional Chinese High-Quality LeetCode Solution Dataset},
  author       = {Ren-Di Wu, Huang Liang Hsun and Twinkle AI community},
  year         = {2025},
  howpublished = {https://huggingface.co/datasets/lianghsun/tw-leetcode},
  note         = {Accessed May 2025}
}

---

License

This dataset is licensed under the MIT License. You are free to use, modify, and distribute the dataset, but please provide appropriate credit to the original authors.