- tokenizer/chat_template.jinja: copied from google/gemma-4-E2B-it-qat-q4_0-unquantized — the export script dropped it, so a host had no way to format prompts.
- eos_token: <eos> -> <turn|>. Gemma 4 ends a turn with <turn|>; <eos> ends the sequence, so a host stopping on it ran past the reply. Matches the Gemma-4-12B/31B bundles in this catalog.
Found by conversion/zoo_verify.py in github.com/john-rocky/coreai-model-zoo.
tokenizer_config: inline the gemma-4 chat template (ships as separate chat_template.jinja upstream, which swift-transformers does not read; without it prompts raw-encode and generation degenerates to turn markers)
tokenizer_config: inline the gemma-4 chat template (ships as separate chat_template.jinja upstream, which swift-transformers does not read; without it prompts raw-encode and generation degenerates to turn markers)
