⚡️ Excited to share another great project built on Fluence!

⚙️ Fine-tuning DeepSeek-R1 for Smart Contract Security by Nishant Pandav.

▫️Nishant fine-tuned DeepSeek-R1-Distill-Qwen-7B for smart contract vulnerability detection using GRPO + LoRA on a single A100 80GB.
▫️ The biggest gain was vulnerable vs clean classification and structured output reliability, improving first-pass screening and integration.
▫️ Fine-grained diagnosis lags, DASP categorization improved modestly, and SWC identification remains weak, limiting audit reliability.
▫️ Training charts show the model learned format compliance faster than security reasoning, improving usability before trustworthiness.
▫️ The Hugging Face release makes the model testable, check out here.
▫️ Built on Fluence GPU for just $30.97, this makes it a strong low-cost task-specific model prototype

📂 Repo on Github
📖 Read more about this case