use 'sagemaker-python-sdk' instead of 'sagemaker' #504
Conversation
jameslamb
added
improvement
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
| - **IAM_ROLE** = Create a new role > Create role | ||
| - **Notebook instance name** = Name of the notebook instance | ||
| - **Notebook instance type** = Type of notebook instance. Select a RAPIDS-compatible GPU ([see the RAPIDS docs](https://docs.rapids.ai/install#system-req)) as the SageMaker Notebook instance type (e.g., `ml.p3.2xlarge`). | ||
| - **Platform identifier** = 'Amazon Linux 2, Jupyter Lab 4' |
There was a problem hiding this comment.
Changes here:
- use the same case as the UI
- remove "IAM Role" (these examples work with the default role added by SageMaker)
- update to
Amazon Linux 2, Jupyter Lab 4(the latest platform SageMaker supports)
| @@ -1,27 +0,0 @@ | |||
| ARG RAPIDS_IMAGE | |||
There was a problem hiding this comment.
There is code in the notebook that writes this... it's not necessary to have a copy checked into the repo too.
| " cupy \\\n", | ||
| " dask-ml \\\n", |
There was a problem hiding this comment.
This was missing, and so the example code failed like this:
Traceback (most recent call last):
File "/opt/ml/code/train.py", line 75, in <module>
train()
File "/opt/ml/code/train.py", line 27, in train
ml_workflow = create_workflow(hpo_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/ml/code/MLWorkflow.py", line 43, in create_workflow
from workflows.MLWorkflowMultiGPU import MLWorkflowMultiGPU
File "/opt/ml/code/workflows/MLWorkflowMultiGPU.py", line 34, in <module>
from dask_ml.model_selection import train_test_split
ModuleNotFoundError: No module named 'dask_ml'
| " flask \\\n", | ||
| " protobuf \\\n", | ||
| " sagemaker\n", | ||
| " rapids-dask-dependency=${{ rapids_version }} \\\n", |
There was a problem hiding this comment.
To ensure dask-ml installation doesn't result in upgrading / downgrading dask and distributed and therefore changing the installed version of RAPIDS libraries.
| }, | ||
| "nightly": { | ||
| "rapids_version": f"{nightly_version}-nightly", | ||
| "rapids_version": f"{nightly_version}", |
There was a problem hiding this comment.
Nothing in the repo relies on this -nightly being there, as far as I can tell.
git grep rapids_versionAnd in fact, this difference is causing some bugs. Look at https://docs.rapids.ai/deployment/nightly/platforms/databricks/#install-rapids-and-dask .... it's saying you should install dask-cuda==25.02-nightly, which does not exist.
There was a problem hiding this comment.
Yeah I think this might be a holdover from a previous naming scheme. Fine to remove it.
jameslamb
changed the title
| }, | ||
| "nightly": { | ||
| "rapids_version": f"{nightly_version}-nightly", | ||
| "rapids_version": f"{nightly_version}", |
There was a problem hiding this comment.
Yeah I think this might be a holdover from a previous naming scheme. Fine to remove it.
| RAPIDS `>24.12` will not be installable on SageMaker Notebook Instances until those instances support | ||
| Amazon Linux 2023 or other Linux distributions with GLIBC of at least 2.28. | ||
| For more details, see https://github.com/rapidsai/deployment/issues/520. | ||
| ``` |
There was a problem hiding this comment.
@jacobtomlinson what do you think about this language for a warning?
I checked the preview and looks like it's rendering OK:
https://rapids-deployment--504.org.readthedocs.build/en/504/cloud/aws/sagemaker/
Co-authored-by: Jacob Tomlinson <[email protected]>
Fixes #443
Contributes to #506
sagemakertosagemaker-python-sdkconda-forge packagessagemakeris just abandoned: Archive this feedstock? conda-forge/sagemaker-feedstock#61sagemaker-python-sdkto avoid fallbacks to older, less-constrained versionsNotes for Reviewers
I tested the SageMaker pages as part of this:
ml.p3.2xlargeinstances, could not provision larger instances likeml.p3.8xlarge(quota limit hit)ml.p3.2xlargeinstances, could not provision larger instances likeml.p3.8xlarge(no capacity)