Chat-Gpt3 Fine-Tuning Questions

Do I get charged for tokens consumed while fine tuning a model? When my model is done, making requests will consume all my training instructions + the request made?

For example:
Training file - Joe ( this tokens) +
request - Whats my name? (this tokens)

Hi Leonardo - welcome to the Forum.

The pricing works as follows: you get charged once for finetuning the model. The amount you pay depends on the volume of your training examples plus the size of the individual training pairs along with a few other parameters. See here for an example of the costs involved (see sub-section “estimate costs”:

Once the finetuned model is in place, you get charged normally for the use of the model as you would for other models. A finetuned model is slightly costlier than the corresponding regular model. I could not find the latest price page otherwise I would have referenced it too.

I hope this answers your question?

The answer given above needs revision.

Model Training Input usage Output usage
gpt-3.5-turbo $0.0080 / 1k tokens $0.0030 / 1k tokens $0.0060 / 1k tokens

Your training file will consist of examples of conversations. All AI-seen language within the file is counted as tokens.

Additionally, the fine-tune is performed in multiple passes called “epochs”, each of which are billed. A typical auto-default for a small file may be eight epochs.

Fine-tune cost = (training file tokens) X (epochs)

Then in use, compare the costs above to normal models. Fine tune model usage is now 2x the cost for input, 3x the cost for output (it used to be an extreme difference):

standard gpt-3.5-turbo: $0.0015 / 1k input tokens; $0.0020 / 1k output tokens

1 Like