Hi, I want to automatically categorize my store categories to Google’s taxonomy in my native language. I found a suitable prompt, but I need to send all 7k categories for it to work correctly. This often results in hitting the TPM limits. Embeddings don’t work because my store’s category names are completely different from the taxonomy. What’s the best way to provide the taxonomy list once and then use it later? Fine-tuning seems like a bad idea since I don’t want to train “how to respond” but just refresh ChatGPT’s memory with proper google tax list.
I guess you managed to fly under the radar! Welcome to the community!
Why do you think this wouldn’t work?
The whole idea behind embeddings is to provide a fuzzy match between ideas - if you pick the right embedding model, mapping meaning from one language to another should work quite well.
1 Like