Your LLaMA model is generating extra text before and after the expected JSON output, and it is not correctly evaluating responsesummary based on the specified factors: relevance and word count

[John6666]

There also seems to be a way to clean up the data assuming that the LLM does not return a complete JSON.

How to request mistral:7b-instruct to skip returning context?

Hi all, my first time around here, so please bear with me if I missed any rules/recommendations. I have been learning to use and build projects using ollama with local model mistral:7b-instruct v3. This particular use case, I am asking the model to impersonate accessibility auditor to generate a report. I have used the model in other contexts as well. Mistral continues to include the full context on the response with thousands and thousands of lines of "[3,1027,781,2744… ", no matter how I req…