paper-implementation
Implement research papers from arxiv. Use when the user provides an arxiv link, asks to implement a paper, or wants to reproduce research results.
When & Why to Use This Skill
This Claude skill automates the end-to-end process of transforming academic research into functional code. By parsing LaTeX source files from arXiv, it extracts complex architectures, loss functions, and hyperparameters to facilitate research reproducibility, code generation, and performance optimization.
Use Cases
- Reproducing Research Results: Automatically implement machine learning models and algorithms directly from an arXiv link to verify paper claims.
- Code Auditing and Verification: Audit existing GitHub implementations against the original LaTeX source to identify discrepancies in hyperparameters or architectural logic.
- Baseline Development: Quickly generate high-quality boilerplate and core logic for new research projects based on state-of-the-art publications.
- Performance Optimization: Profile research code and implement custom CUDA or Triton kernels to scale academic prototypes for production-level compute.
| name | paper-implementation |
|---|---|
| description | Implement research papers from arxiv. Use when the user provides an arxiv link, asks to implement a paper, or wants to reproduce research results. |
Paper Implementation
Input
Require arxiv link (not PDF). Fetch LaTeX: https://arxiv.org/e-print/XXXX.XXXXX (.tar.gz with .tex files)
Phases
1. Discovery
Parse LaTeX for: architecture, algorithms, hyperparameters, loss functions, datasets. Search existing GitHub implementations. Note ambiguities.
Gather: Scope (train/inference/finetune)? Scale (model size, compute)? Baseline codebase? Priority (accuracy/speed/memory)? Validation method?
2. Verification
If repo exists: audit against LaTeX source. Check architecture, hyperparameters, training procedure. Identify discrepancies.
3. Refinement
Present findings, ask questions, iterate until execution steps are perfectly clear.
4. Implementation
Build/modify code. Write correctness tests. Profile performance.
5. Optimization (optional)
Profile with nsight-systems/torch.profiler. Write custom CUDA/Triton kernels. Benchmark with measurements.
Compute Strategy
- Local 3090: quick tests, debugging, small validation
- VP H100s: training runs, large experiments (ask before provisioning)
On arxiv link
- Parse LaTeX source
- Search existing implementations
- Present structured questions
- Wait for answers, refine, then proceed autonomously