PSA: Gemma 4 12B is NOT completely broken for coding and tool calling, you need a special chat template
A Reddit user has shared a workaround for issues with Google's Gemma 4 12B model, specifically concerning its tool-calling capabilities. The problem, which caused failures in coding tasks and harnesses like OpenCode, can be resolved by using a specific chat template file. This fix, available via a GitHub gist, requires compiling llama.cpp from source and then running the model with the custom template file, enabling a more accurate evaluation of its coding abilities. AI
IMPACT Enables more accurate evaluation of Gemma 4 12B's coding capabilities by fixing tool-calling bugs.