Gemma 4 Crashes Llama.cpp on Images — And the Sneaky Fix
Loading Gemma 4 into llama.cpp for image tasks? Expect a brutal crash. One ubatch tweak saves the day, but why's this still a headache in 2024?
⚡ Key Takeaways
- Gemma 4 vision needs explicit ubatch 2048+ for non-causal image tokens.
- Cap tokens at 1120 max; tiered budgets prevent overkill.
- Llama.cpp crash fix: simple flags, but exposes multimodal growing pains.
Worth sharing?
Get the best Developer Tools stories of the week in your inbox — no noise, no spam.
Originally reported by dev.to
