Code Model Security Flaws and MoE Fine-Tuning Fragility

Research indicates that code models are inherently unsafe due to public training data, and simple safety fine-tuning is insufficient. Furthermore, MoE models are more fragile than Dense models during SFT, showing high sensitivity to hyperparameters and routing instability.

2026-08-12 ~ 2026-08-12 · 2 related posts

• MoE is Fragile During SFT; Coding Scaling Laws Vary Drastically Across Languages — mdancho84 · 2026-08-12
• Code Models Are Insecure by Default; MoE is More Fragile Than Dense Models During SFT — mdancho84 · 2026-08-12