Settings

Theme

Gemini Distillation Service

docs.cloud.google.com

20 points by asawfofor 9 days ago · 11 comments

Reader

moralestapia 9 days ago

Fine-tuning and distillation used to mean two different things. Now not so much anymore, I see.

  • bizzletk 9 days ago

    Can you explain the difference?

    • porridgeraisin 9 days ago

      Distillation originally meant matching the distribution of the student model to the teacher model using something like a KL divergence.

      When you instead fine-tune the student on the samples from the teacher, which is what people mean by distillation today, you are in effect doing a monte-carlo version of the same thing. While in theory this is higher variance, given modern setups where the student and teacher are both large and are RLd heavily (leading to a sharp teacher distribution), and given that you typically use lots and lots of samples, it ends up OK.

    • nerdsniper 9 days ago

      Fine-tuning is done against a dataset. Distillation is done against a model.

      • verdverm 8 days ago

        Distillation is used to build part of a data set for fine-tuning (loosely interpreted). Advanced model traces are useless if you don't have a base model that is good enough to be improved by them.

  • sapienskid 8 days ago

    How to get access?

rs38 8 days ago

404?

Keyboard Shortcuts

j
Next item
k
Previous item
o / Enter
Open selected item
?
Show this help
Esc
Close modal / clear selection