Why does the chat template include reasoning/thinking components?
#45
by usrlocalben - opened
Although it is not said explicitly in the model-card nor tech report, this doesn't appear to be a reasoning /CoT model. Recent changes in llama.cpp cause it to emit all of its output in a <think> block (reason unclear, there's some churn since the new chat parser) -- but I wonder: Why does the chat_template have reasoning treatment at all?
Can you clarify: is it expected to have a reasoning phase? Should it be run with enable_thinking explicitly set to false?