W2 lab for ungraded people - offline trained model

Course Q&A Generative AI with Large Language Models
1

Hi, I’m trying to follow the walkthrough of the lab that i can’t access because I’m doing the ungraded version. I want to know if there is a way to download the offline trained model so I can follow the exercise. I’ve tried to fine tune the model on my computer but it will need to much time or may be it’s impossible.

2

I am not a mentor for that course, but i expect that training requires a lot of GPU power.

3

Graded labs were designed to run with the Vocareum environment. Jupyter Notebook is loading in ml.m5.2xLarge 8vCPU+32GiB (amazon SageMaker notebook). Maybe It can not load in locally.

1 Like
4

I know that the lab run on a powerfull ambient. Is for this that I asked the tuned model that they prepared offline. I found on hugginface other model tuned for summarization but I can’t use it to make the comparison because the index of the dialogue don’t match. I will try another way

1 Like
5

I ran the lab for week 2 on my desktop - that does not have a GPU - and it works fine, I went a bit overboard and had to let it run overnight but overwise it works fine.

1 Like
6

Hi the notebook downloads the model checkpoint you are talking about from AWS S3 . You need to have the AWS CLI installed locally for this to work on your machine.

Look for the line in the notebook with a similar command (modified to preserve intellectual property):

!aws s3 cp --recursive s3://<...>/flan-dialogue-summary-checkpoint/ ./flan-dialogue-summary-
checkpoint/
1 Like
7

Hi the notebook only performs training for max_steps=1, to keep infrastructure costs manageable. The model checkpoint (trained on a GPU instance for some time) is then loaded from an external store (AWS S3).

1 Like
8

Hi, what are the credential values to have access to this model? I get this error running the cell offline “fatal error: Unable to locate credentials”

1 Like