Fix: Hard-code chat_template into tokenizer_config.json

Upload merged model (Base + LoRA)

initial commit