mistralai/Mistral-7B-Instruct-v0.3 · An stable Output with Mistral-7b-Instruct-v0.3

I am Working with Soc2 report analysis, in that i want an output as socring of five trust service criteria (Security, Availability, Processing Integrity, Confidentiality and privacy). a model can give score based on compleeness: how well the report address all trust service criteria and associated controls.

The Mistral-7B-Instruct-v0.3 model give different output even if prompt and the input report text/ data is same.

if any one have solution for the smae than please help me to solve the issue.

Thank you