Need Gemini LLM Output to be more grammatically tuned - Google Cloud Community
Share this topic
LinkedIn
Twitter
Copy URL
Issue details:
We are using Gemini Flash 002 for image and video analysis to predict content details. The output is in the form of tags that we compare against a fixed set of vocabulary provided by our business.
For example, if there is an image of a table with food items, the LLM Output might be:
{ Items: "Apple ", "Beverage", "Soups" }
However, one challenge we are encountering is that sometimes the LLM outputs singular forms of an item even when there are multiple quantities. For instance, it may output "soup" instead of "soups" or "beverage" instead of "beverages".
We are striving to ensure that the output always returns plural forms, even when singular quantities are present. We have attempted various adjustments to prompt the system, but have not had success thus far.
If you have encountered a similar issue or have any suggestions for resolving this, please share your insights.
