feat: add rhoai 3.2 fine tuning examples#27
Conversation
briangallagher
force-pushed
the
add-3.2-finetuning-examples
branch
5 times, most recently
from
63d2581 to
6fd60d9
Compare
Signed-off-by: Brian Gallagher <briangal@gmail.com>
briangallagher
force-pushed
the
add-3.2-finetuning-examples
branch
from
6fd60d9 to
7cbcdce
Compare
Fiona-Waters
left a comment
Fiona-Waters
left a comment
There was a problem hiding this comment.
Some small nits after code review, otherwise they look good.
kramaranya
left a comment
kramaranya
left a comment
There was a problem hiding this comment.
Thanks @briangallagher!
Looks good to me, just left a few comments
| * Log in, then go to _Data Science Projects_ and create a project: | ||
|  |
There was a problem hiding this comment.
Shall we update those images with RHOAI 3.2?
There was a problem hiding this comment.
ok if we fix this for 3.3? Given time constraint? We don't currently have a 3.2 cluster available. 3.2 fast release also.
There was a problem hiding this comment.
sure, I think it should be fine
| "!python3 -m pip install datasets transformers accelerate bitsandbytes huggingface_hub\n", | ||
| "!python3 -m pip install --force-reinstall --no-cache-dir -U datasets\n", | ||
| "!python3 -m pip install --force-reinstall --no-cache-dir -U \"kubeflow @ git+https://github.com/opendatahub-io/kubeflow-sdk.git@v0.2.1+rhai0\"\n", | ||
| "!python3 -m pip install --force-reinstall --no-cache-dir -U training-hub==0.4.0\n", |
There was a problem hiding this comment.
Just thinking out loud, why training-hub is not a dependency in kubeflow sdk?
There was a problem hiding this comment.
iirc, Saad had an opinion on this that training-hub was too "big and complex" to be a dependency of the sdk. We could make it an extra but then we have to consider how to deal the training-hub extras such as cuda. I think it makes sense to keep separate.
| "!python3 -m pip install --force-reinstall --no-cache-dir -U datasets\n", | ||
| "!python3 -m pip install --force-reinstall --no-cache-dir -U \"kubeflow @ git+https://github.com/opendatahub-io/kubeflow-sdk.git@v0.2.1+rhai0\"\n", | ||
| "!python3 -m pip install --force-reinstall --no-cache-dir -U training-hub==0.4.0\n", | ||
| "!python3 -m pip install --force-reinstall --no-cache-dir -U ipykernel" |
There was a problem hiding this comment.
why do we need ipykernel?
There was a problem hiding this comment.
hmm, I'm struggling to remember now. I did hit a an issue that required me to install it but struggling to remember. I think it's maybe safer to leave in and revisit for 3.3?
There was a problem hiding this comment.
Its a dev dependency in training-hub and mini-trainer. I had to add it for e2e tests also. BUT it is included in the universal image so when we are using that we won't need to pip install separately.
There was a problem hiding this comment.
Why do we need those changes in this file?
There was a problem hiding this comment.
It was failing format or lint. I've shared this with Shri and he was fine with us committing
Signed-off-by: Brian Gallagher <briangal@gmail.com>
|
Thank you @briangallagher! |
Signed-off-by: Brian Gallagher <briangal@gmail.com>
3 participants
Add RHOAI 3.2 fine tuning examples for OSFT and SFT