README.md

Community Results & Feedback

Genesis AI Code is designed for real-world training.
We strongly encourage developers and researchers to share empirical results when using this dataset.

If you train, fine-tune, or evaluate a model with this dataset, please consider reporting:

What to Share

• Base model used
• Training method (SFT, LoRA, QLoRA, full finetune, agent training, etc.)
• Which split(s) were used:
  • Instruct
  • Thinking
  • Reasoning
• Evaluation method (benchmarks, internal tests, SWE-bench, CI pass rate, etc.)
• Observed improvements or regressions
• Notable behaviors (planning quality, verification discipline, safety improvements)

Where to Share

• Open an Issue or Discussion in this dataset repository
• Link to:
  • training logs
  • benchmark tables
  • evaluation scripts
  • demo videos or repos (if public)

Why This Matters

Genesis is built to evolve based on real usage, not assumptions. Community-reported results directly influence:

• future Genesis updates
• new training signals
• improved reasoning and safety patterns

Even partial results or informal observations are valuable.

Minimal Report Template

Model:
Training method:
Dataset split(s):
Compute budget:
Evaluation method:
Observed changes:
Notes / caveats: