Builder Feedback: Does our Solidity v0.8.28 audit-to-remediation format fit your QLoRA / RLVR reward harness?
#3
by Offlin33er - opened
Hi smart-contract & LLM security builders! π
Over 120 teams have downloaded this 100-row preview sample. We built the full 6,000-row v4 multi-domain corpus with zero public Hugging Face overlap, specifically targeting modern Solidity v0.8.20β0.8.28 DeFi vulnerabilities (ERC-4626 vault inflation, L2 bridge sequencer downtime, Account Abstraction gas griefing).
If you've run this sample through an eval pipeline or fine-tuning run:
- Audit vs. Patch pairing: Is the side-by-side vulnerable code + reasoning rationale + hardened patch structure working well for your loss functions or RLVR verifiers?
- Compiler & Tooling: Did you encounter any issues compiling the remediations with Foundry/Hardhat?
- Targeted vulnerabilities: What specific attack vectors in 2026 DeFi protocols do you want deeper coverage on for future splits?
Please drop your unvarnished feedback below!