I'd be interested to see how well these fine-tuned models compare to Claude 3 Haiku (or one of the more expensive Claude models) with a larger set of examples.
The Claude models all have a 200,000 token limit and respond _really_ well to examples - you can feed them in as chat JSON message pairs of user input / ideal assistant output.
Haiku is dirt cheap for this kind of thing and with 200,000 tokens you can probably provide a dozen or so examples.
Comments
I'd be interested to see how well these fine-tuned models compare to Claude 3 Haiku (or one of the more expensive Claude models) with a larger set of examples.
The Claude models all have a 200,000 token limit and respond _really_ well to examples - you can feed them in as chat JSON message pairs of user input / ideal assistant output.
Haiku is dirt cheap for this kind of thing and with 200,000 tokens you can probably provide a dozen or so examples.