The TODO list is awesome
Let me tell you a big thank you!
For learners and newbies (like me) the fact that you are updating the TODO list in the model card is a very good way to see what it takes to build these GGUF.
❤️
He's smart, otherwise he knows that his avid followers like me will not leave him alone, and will keep asking him every minute about it 😉
thanks!
i have an outdated quant cookers guide here too for folks looking to learn the ropes or cook their own custom quants: https://github.com/ikawrakow/ik_llama.cpp/discussions/434
the TODO helps keep my scattered brain focused too lol
On another note, which chat template are folks using for GLM-4.6?
On another note, which chat template are folks using for GLM-4.6?
Probably depends on how they are using it. You can add --jinja to use the included jinja template for openai api compliant endpoints via /v1/* stuff. I've been using it without specifying anything and it seems to work okay and adding /nothink at the end of my prompt makes it skip thinking for a quick response so far using chat completions endpoint.
Folks on the beaver ai club discord tend to use silly tavern and custom templates for more flexibility and injecting thinking cots etc for more advanced usage with the text completions endpoint.