I would expect so! I'm currently running Gemma 4 E4B evals and it's behaving the same. Better with guardrails. There might be a floor where any error nudge confuses the model more than helps, but I haven't found it across many 8B families and now Gemma 4 E4B.
I would expect so! I'm currently running Gemma 4 E4B evals and it's behaving the same. Better with guardrails. There might be a floor where any error nudge confuses the model more than helps, but I haven't found it across many 8B families and now Gemma 4 E4B.