Hi everybody, a couple of newbie questions: 1) Is it possible to use a custom model for domain-specific translation purpose, like translating medical concepts from one language to another by feeding a chatgpt 3.5 model (from my inormation 4th version can not yet be customized) with translation memories in form of pairs of source_language_concept - target_language_concept as jsonl-formatted files? From your experience does this improve the accuracy of the model afterwards? is there any example of such project available in the community or on the web? 2) What would be the openai licensing model suitable for this, is the “Plus” upgrade from the free license suitable or does it require a higher-price model? Thanks everybody and looking forward to your helping hand.
Welcome to the OpenAI community @fb4docs
Here’s a sample before fine-tuning
The gpt-3.5 and gpt-4 models will be available for fine-tuning later this year.
The Plus is a subscription plan for chat.openai.com.
If you want to use the API api.openai.com, you have to add a payment method, and the billing occurs monthly on a pay-as-you-go basis for the tokens consumed per pricing
Thanks for the info. Is this a model you’ve already fine-tuned on a lower-than 3.5 model? You mention that fine-tunning is not yet available on 3.5 and above yet.
Hello!
Please tell me how to create data for fine-tuning using a thematic dictionary of 40,000 entries?
Hi @saharin
Here’s OpenAI’s guide to preparing a fine-tuning dataset
Can you elaborate more on this?
We want to add translator functionality to the English-Ukrainian military dictionary website, which works using the GPT model trained on dictionary data, with a user interface similar to Google Translate.
GPT comes pre-trained.
Before you proceed for fine-tuning, you should check if the model already achieves what you are looking for using playground