Use AWS Bedrock to train own private models per user, so each can have his own private trained models, is this possible?

0

Hi! I would like to know if with the AWS Bedrock Java SDK I can create a fine tuning model per user that will be trained periodically by my system behind the scenes, so each user has his own private models and later when the user interact with the custom chatbot built in my app using behind the scenes AWS Bedrock, the chatbot can give each user their own personalized responses learned from their own trained private models. Is this possible? Also do you know how much time it will take for AWS to release Bedrock in european regions?

Thank you guys!

cao95
已提問 7 個月前檢視次數 471 次
2 個答案
0

Hi,

Quite interesting! It is possible to create custom models and submit fine-tuning jobs using APIs. There are a few examples here using Python SDK, there should be similar capabilities using Java SDK.

Please consider the models which are eligible for fine-tuning and the provisioned throughput aspects as described here.

You can control the access to relevant model endpoints using api-gw and lambda so that the users use the corresponding model endpoints.

By the way, your use-case also seems to be a good fit for Retrieval Augmented Generation (RAG), which could be used in select cases as an alternative to fine-tuning. Please check this blog.

Regarding the availability of Bedrock in Europe, I cannot provide you an estimate. I suggest checking here.

Thanks!

profile pictureAWS
Rama
已回答 7 個月前
  • Hi Rama!

    Thank you so much for the answer, it helped me a lot!

    I have one more question, I don't see clearly the "provisioned throughput" thing. I would need to purchase that in order to be able to use my own trained model?

    Thank you!

  • Hi, I am glad the answer helped! Yes, this is correct, you would need to purchase provisioned throughput in order to deploy a custom fine-tuned model for inferencing. Thanks, Rama

  • And this is per model? Let's say that I have 2 users in my app and I want to create one model per user to train each chatbot with his own data. I would need to purchase 2 provisioned throughput, one per user?

    Thanks!

  • Hi, The provisioned throughput depends on the model units consumed, I suggest get in touch with your AWS account manager he will help you to get a firm estimate. You can also refer to the examples on this page https://aws.amazon.com/bedrock/pricing/

    Meanwhile, Amazon Bedrock is now available in Europe Frankfurt region. https://aws.amazon.com/about-aws/whats-new/2023/10/amazon-bedrock-europe-frankfurt-aws-region/

  • Thank you!

0

Hi,

I would also suggest checking this blogpost, which describes the similar idea: https://aws.amazon.com/blogs/containers/build-a-multi-tenant-chatbot-with-rag-using-amazon-bedrock-and-amazon-eks/

profile picture
專家
已回答 7 個月前
  • Thank you!

您尚未登入。 登入 去張貼答案。

一個好的回答可以清楚地回答問題並提供建設性的意見回饋,同時有助於提問者的專業成長。

回答問題指南