-
Notifications
You must be signed in to change notification settings - Fork 1.2k
Open
Description
The list of inference_graviton image URIs is out-of-date, making it difficult to deploy models on graviton with the latest PyTorch version.
The SDK says 2.4.0 is latest version:
sagemaker-python-sdk/sagemaker-core/src/sagemaker/core/image_uri_config/pytorch.json
Lines 1352 to 1363 in ad190b9
| "inference_graviton": { | |
| "processors": [ | |
| "cpu" | |
| ], | |
| "version_aliases": { | |
| "1.12": "1.12.1", | |
| "2.0": "2.0.1", | |
| "2.1": "2.1.0", | |
| "2.2": "2.2.1", | |
| "2.3": "2.3.0", | |
| "2.4": "2.4.0" | |
| }, |
Although 2.6.0 has been available for a year:
- https://github.com/aws/deep-learning-containers/blob/master/available_images.md#sagemaker-framework-arm64graviton-containers-sm-support-only
- aws/deep-learning-containers@b41cd39
Workaround
It's still possible to manually provide the image URI to the ModelBuilder which leads to a working deployment.
Metadata
Metadata
Assignees
Labels
No labels