More fun with #AI to start my day. A buddy of mine has a some web forum data where doctors and nurses answered people's questions: 20 years worth of Q&A. He asks me how he might train an AI model on that data. I have a vague idea and tell him "you can fine tune a foundation model with that data." To save myself typing I ask #AmazonQDeveloper :
I have a bunch of posts on a web forum stored in a SQL database...I want to use Amazon Bedrock to create a chatbot that answers questions using a fine-tuned model. Write out the high-level steps I will need to take.
(I said a bunch of stuff in that ellipsis about what the data was)
Among the many steps it tells me to do, it says:
Format your data according to #Anthropic's fine-tuning specifications for #Claude
That's interesting. So I ask: where can i get more information on Anthropic's fine-tuning specifications for Claude?
It replies:
Amazon Q is having trouble responding right now:
0: unhandled error (InternalServerException)
1: service error
2: unhandled error (InternalServerException)
I chuckle, press the up arrow, and resubmit my query. It replies:
I apologize, but I need to correct my previous response. I made a mistake in suggesting that Claude could be fine-tuned through Amazon Bedrock. Currently, Claude models (including Claude Sonnet) on Amazon Bedrock are not available for fine-tuning by customers.
Great. Thanks, Q. It's like working with a chipper intern.