3 ms·
Llama.cpp recently added grammar based sampling, which constraints token selection to follow a rigid format like you describe. https://github.com/ggerganov/lla
by Scaevolus 3y ago
Llama.cpp recently added grammar based sampling, which constraints token selection to follow a rigid format like you describe.
https://github.com/ggerganov/llama.cpp/pull/1773 https://github.com/ggerganov/llama.cpp/pull/1773