OpenAI releases updates to fine-tuning and custom models

Community
1

Rolling out today, new features for developers to enhance fine-tuning control, including epoch-based checkpoint creation, a comparative Playground UI for model evaluation, third-party platform integration (beginning with Weights and Biases), comprehensive validation metrics for better quality insights, hyperparameter configuration via Dashboard, and fine-tuning dashboard enhancements for deeper metric visibility and job reruns.

2 Likes
2

My favorite feature in this update is the compare section, found here:
The image displays a user interface with a vertical navigation menu on a dark background, highlighting the "Compare" option as selected. (Captioned by AI)
Because not only do you get to see the results:

The image shows a side-by-side comparison of two chat interactions with dialogue boxes where the system quips "You are a math genius," followed by a user asking "What is 2+2?" and the assistant on both sides correctly answering "2 + 2 = 4," labeled as gpt-3.5-turbo on the left and gpt-4-turbo-preview on the right. (Captioned by AI)
The image shows a side-by-side comparison of two chat interactions with dialogue boxes where the system quips "You are a math genius," followed by a user asking "What is 2+2?" and the assistant on both sides correctly answering "2 + 2 = 4," labeled as gpt-3.5-turbo on the left and gpt-4-turbo-preview on the right. (Captioned by AI)2102×381 22.6 KB

You can also see the latency between the two models to help you decide which model you want to use if your project is very dependent on quick responses.
The image is mostly dark with two lines of white text stating "Latency 758ms - Finish reason: stop" on the left and "Latency 906ms - Finish reason: stop" on the right. (Captioned by AI)
The image is mostly dark with two lines of white text stating "Latency 758ms - Finish reason: stop" on the left and "Latency 906ms - Finish reason: stop" on the right. (Captioned by AI)1309×30 2.87 KB

(Side note: Having access to setting your fine-tuned models names is super important for me as it will help me not loose track of which model is trained on what.)

1 Like
3

Already had been doling info about it.

2 Likes