Degraded response quality on v 0.1.33

### What is the issue? I have an application that creates summaries of text and noticed that - after upgrading to v0.1.33 - the quality of the generated content was MUCH worse, without any change in the code or model. - in some summaries it seemed that the model would not respect instructions, and would be extremely verbose - in some cases, the model seemed to hallucinate , and provide opinions that did not reflect the content that was summarized. I rolled back to 0.1.32, and the responses immediately went back to normal. This is running on docker, in a UBUNTU 22.04 VM with 128G RAM (no gpu), and the model I have been using is dolphin-mistral:v2.6. Anyone else experienced something similar? It "looks" like Ollama was using a completely different model (that machine has a few installed) or that it wasn't allocating enough resources for the model to perform (these are just impressions) ### OS Linux ### GPU Intel ### CPU Intel ### Ollama version 0.1.33