Llama.cpp reasoning_content

#8
by brianw - opened

Are template changes necessary to get llama.cpp to output reasoning content in the "reasoning_content" field of responses, or simply a setting that needs to be set?

Use the original GLM-4.5-GGUF chat template to get reasoning parsing to work properly. I don't know what new changes unsloth found were needed, but what ever it fixed also broke this feature.

No dice. It seems like the original GLM-4.5 template also doesn't output any reasoning_content.

https://github.com/ggml-org/llama.cpp/pull/15186

use above link PR to get it working.

Sign up or log in to comment