rh-aiservices-bu / insurance-claim-processing Goto Github PK

so dry-run people don't attempts stuff we know is still half baked.

Create the content for Section 02 (Connection and Setup)

To be on the safe side. If people leave too many NB opened, kernel will crash on heavy workloads (model retraining).

If people launch multiple trainings, or do something weird, space will be filled on workbench. Better be on the safe side.
As we have switched to Ceph-RBD, this would in fact not consume more than necessary.

To avoid getting 50 pods an hour later while still in the workshop

For the app deployment/update part, it would be better to use a VSCode image in the cluster.

ensure that we have some gitops magic that sets up the VSCode image as a custom image.

this is weird but not a complete showstopper:

when you click to select, it does not look like it worked.

There are only the settings, but not where to create the workbench itself.

Create the content for Section 04 (Image Recognition)

it should exist in this folder: https://github.com/rh-aiservices-bu/insurance-claim-processing/tree/dev/lab-materials/04/images

ensure that the following images are pre-pulled onto the nodes:

• pytorch notebook image
• modelmesh image
• others?

use variables for things like

• RHOAI
• Full prod name
• Lab/Workshop etc

section 03 will have a lot of different files.
can we move the current content of https://github.com/rh-aiservices-bu/insurance-claim-processing/tree/dev/lab-materials/03 into a subfolder called 07 to match the instructions numbering? (https://rh-aiservices-bu.github.io/insurance-claim-processing/modules/dev/03-07-sanity-check.html)

After each main section, notebook kernels should be shut down to free memory.

Image recognition should be a bit faster. Keep the content but make more of it optional

remove all S/M/L option sizes.

replace with a single size for each.

Call it "lab-sized" or something like that.

check with Tony.

Can fail at "quality test".
Can either lessen the requirement to succeed or try a few times and average.

we should copy this image out of @RHRolun 's quay account and into the aiservices account, and then update the instructions to use that instead.

since there are no built-in safeguards beyond the ones built into the prompt, it's easy to get the LLM to say very offensive things. (unlike chatGPT, for example).
Should we warn the users, and have a legal disclaimers that with or without modification of the prompt, we can't be held responsible for what mistral will spit out?

also. don't go sell an insurance solution.
This is not "what's in openshift AI". This "what a customer could build with openShift AI".

Send something small instead

Create the content for Section 06 (Productization)

and then update the instructions to use that one.

review instructions in section 4 and update notebooks to match the overall flow

• How to properly apply the YAML?
• Update Doc
• Pipeline failing

this happens on the "response quality" check.

happens to 5% of users, and passes on second try.

Check that we always refer to elements with the same names:

• "OpenShift Console"
• "OpenShift AI Dashboard"
• "Workshop" (instead of Lab)

Create the content for Section 05 (App Deployment)

pre-pull images for pipeline server just like in:
https://github.com/rh-aiservices-bu/rhoai-rh1-testdrive/blob/main/bootstrap/rhoai-config/images-puller.yaml#L70

Create the content for Section 01 (Background)

Images are not the same on the RHS and and the original claim content.

Screenshots:

Description does not say where to click to create a Data Connection. Although it may seem obvious, beginners may miss it.

Describe the bug
Clicking here:

breaks things:

To Reproduce
Steps to reproduce the behavior

Expected behavior
A clear and concise description of what you expected to happen.

Screenshots
If applicable, add screenshots to help explain your problem.

Additional context
Add any other context about the problem here.

The % shown in the two images differs, mostly because of a preprocessing bug.

copy from Test-drive workshop

so that in the future, RH people can book this env and go through this lab themselves.

Screenshot shows Python 3.9 but the Workbench we use is 3.11

Top right of content, same line as breadcrumb

Create the content for Section 07 (End of Lab)

Create the content for Section 03 (LLM)

for each user (user1 - userX):

• create a project called "userX-auto"
• ensure userX is admin of that project
• deploy minio
• deploy the DS pipeline server
• create a workbench
• git clone the project

then

• adjust instructions/code to match
• keep the instructions where people create their own artifacts, but instead call it "userX-diy"