Tuning screen
- Print
- PDF
Tuning screen
- Print
- PDF
Article Summary
Share feedback
Thanks for sharing your feedback!
Available in Classic and VPC
This is a description of the tuning screen composition. Tuning is a space where you can learn and use the HyperCLOVA language model of CLOVA Studio JP in an optimized form for the type of work and language you want. The benefits of using tuning include:
- Playground has a token restriction (2048 tokens), but tuning allows more diverse examples without restrictions
- You can configure the dataset and output it in any way you want.
- You get a lighter model, shorter task execution time, and cheaper costs.
- Performance can be improved in congested sections where prompts don't lift performance
Note
The current tuning outputs the result based on the following parameters. The output results can vary even with parameter adjustments, so create a test app and adjust the settings after training. We're going to make improvements so the tuning model can be imported and used in Playground.
Parameter | Value |
---|---|
Top P | 0.8 |
Top K | 4 |
Maximum tokens | 300 |
Temperature | 0.85 |
Repetition penalty | 5 |
Tuning screen
The tuning screen is laid out as follows.
- [Create]: Click to create a job optimized for that type of job
Was this article helpful?