Zynerji's picture
Inline chat_template into tokenizer_config.json — llama.cpp GGUF conversion reads it from here, not from chat_template.jinja, so quantised builds were shipping with no chat template
401daec verified