Raghu/baseimage update#4134
Conversation
Test Results for assets-test0 tests 0 ✅ 0s ⏱️ Results for commit 5a4ddaa. ♻️ This comment has been updated with latest results. |
Test Results for training-model-mgmt-unittests29 tests 28 ✅ 7s ⏱️ For more details on these failures, see this check. Results for commit 5a4ddaa. ♻️ This comment has been updated with latest results. |
|
|
||
| code: ../../src/ | ||
| command: | | ||
| pip install transformers==4.48.0 torch==2.2.2 numpy==1.23.5 pandas==1.5.3 urllib3==1.26.19 --no-cache-dir --force-reinstall |
There was a problem hiding this comment.
runtime dependencies installation? sweet...
There was a problem hiding this comment.
I initially tried moving dependencies into requirements.txt, but ran into an issue:
[module 'azureml.evaluate.mlflow' has no attribute 'hftransformers']
This happens because the required transformers and compatible torch versions aren’t guaranteed to be installed or active early enough in the environment lifecycle.
To ensure compatibility and avoid evaluation failures, I opted to install the runtime dependencies explicitly in the spec.yaml. This guarantees they’re available at execution time, avoiding any cached or conflicting versions.
There was a problem hiding this comment.
I incorporated the required libraries directly into the Dockerfile rather than using the spec.yaml file.
After adding those changes, I ran import_model, it is now running successfully.
There was a problem hiding this comment.
@vizhur, Could you please review this PR
Included the necessary libraries directly in the Dockerfile instead of using the spec.yaml file.
2 participants
Tested locally It is working,