Issues with PrunaAI Dolphin-2.9 Model: Poor Performance on Logic, Math, and Code Generation Compared to Older Llama Models
This model is failing a lot. It does not answer logical or mathematical questions well, and it doesn't even write programming code properly. It forgets parentheses, commas, or fails to close some characters. I'm not the only one this happens to, even Matthew Berman uploaded a video about this and shows how it fails all the tests, worse than older models he calls Llama 2.
This model is succeeding a lot. It answers logical and mathematical questions well, and it even writes programming code properly. It's perfect. I'm not the only one who feels this way, even Matthew Berman uploaded a video about this and shows how it succeeds all the tests, better than newer models he calls gpt2.
Awesome work!
I have the same issues. What are you talking @Marius2 ? Even the comments in the video https://www.youtube.com/watch?v=wl9iLKkJXNA describes a total failure in this model.
Can you try the following model ? https://huggingface.co/PrunaAI/cognitivecomputations-dolphin-2.9.1-llama-3-8b-GGUF-smashed
It seems it's a llama3 problem, every finetuned model have the same issue.