Needs more work

#2
by nenor - opened

I have tested your model and omg it is so dumb and doesn't follow conditions, i asked it if there are doublons and i saw 10 doublons and it checks so quick and replies "no there isn't"....

It doesn't stop when i told it to stop per turn. it doesn't even think large about a project it just acts like every qwen just saying 2% of the work then if you add something it says "oh yeah i thought about it".. BRUH

and it says in loop creating the same file over and over

I don't mean to be rude, but based on your previous comments on other models, you seem to be running into tool call issues very often. Are you sure it is the models fault and not an issue with the way you are running the model? It would be helpful if you could share more details about your setup. Are you using high or max effort thinking supported by the chat template? Are you using the recommended generation parameters: https://huggingface.co/pragmaticcs/PentaCoder-9B-GGUF#recommended-generation-parameters? Are you using f16 kv cache (or at least q8)?

Sign up or log in to comment