I created a prompt for generating a color palette based on the user’s query. For example, consider the color palette for Independence Day. Based on this, I created 10 training examples and fine-tuned the model. This model worked perfectly for Independence Day, but it did not perform well for other categories of color palettes, like festivals. So, we created 10 training examples for festivals and updated our model.
However, this model does not provide the proper colors for Independence Day as the previous model did.
Colour pallets only have a single correct set of values, this makes it data and not a “style” or a way of processing other data.
You might look into embeddings, which are a way of storing text and other related data in a semantic way, i.e. the meaning of the text can be searched, if that was combined with traditional word searches you may produce a more accurate system.
Hi @Foxabilo in my case I have the algorithm to create the color palette by Chat Gpt and not give all data of the color palette
My issue is that we need to support continuous updates of the fine-tuned model. but in this update previous model working properly. like in my case base fine-tuned model works properly but when I update this model it does not work for the previous train data. this is the issue
Unfortunately not every use case is viable, your current approach seems to have reached a dead end. It may be time to try other methods. Fine tuning is not the correct mechanism to use for this task.
Let me know which method should I prefer for my below task.
my requirement is to create a color palette for the poster that has 4 colors and this is based on the poster’s details. What should I prefer for that? also there is not any fic category design there are several category.