Small Gemma 4 tokenizers were recently updated by Google - fix for an issue that caused nonsense output when encountering certain tokens
retained text
Gemma 4 E4B as an example, but the same fix was applied to E2B and 12B too. Not sure if this will trickle down to GGUFs
document fetched 2026-11-17 00:42 · canonical https://reddit.com/r/LocalLLaMA/comments/1xxxs0a/small_gemma_4_tokenizers_were_recently_updated_by