I’m considering fine-tuning the GPT-3.5-Turbo model, which has a current cost of $0.50 per 1M tokens for input and $1.50 per 1M tokens for output. After fine-tuning, will the cost per token increase when I use the fine-tuned model for chat or other API interactions? Or does it remain the same as the base model pricing?
1 Like
2 Likes
Thank you for the response. Also, can you tell me which pricing to consider for each model after finetuning.
1 Like
Hi @neel1 - not sure I fully understand the question. Under the link above you can see the prices for the use of fine-tuned model for the different models that can be fine-tuned.
Is it this what you are interested in?
1 Like
Yes, I have few questions regarding this:
- In this table are the rates referring to the use of these models after fine tuning? If no, then what will be the usage rates?
- Also what does the training tokens here mean?
- What token rate is considered when I am fine tuning any model (for eg: 3.5-turbo)
1 Like
1.- During and after
2.- The tokens you want to feed the machine to be fine-tuned (training)
3.- for the tuning itself consider the training price
2 Likes
In addition to @samuel.cortes.arg’s comment, just one additional point:
Input tokens refers to the token as part of your prompt (i.e. the system message (if applicable) and the user message) while output tokens essentially refers to the amount of tokens generated by the model response.
Let us know if anything else needs clarification.
2 Likes
One last question:
The rates here, in the image that I attached, will be taken into consideration when I am using the basic model (not fine-tuned). While the rates in the ‘fine-tuning models’ section will be taken into consideration after I fine-tune this model. And during the fine tuning process the ‘training tokens’ rate will be taken into consideration.
Is that correct?
1 Like
correct. these are the rates for the regular model.
1 Like
Understood. Thanks a lot @jr.2509 and @samuel.cortes.arg
2 Likes