This example demonstrate how to run distributed batch scoring job in TensorFlow on Azure Batch AI cluster of 2 nodes. Inception-V3 model and unlabeled images from ImageNet dataset will be used.
- For demonstration purposes, pretained Inception-V3 model and approxinately 900 evaluation images from ImageNet dataset will be deployed to Azure Blob Container
- Standard output of the job will be stored on Azure File Share;
- Azure Blob Container and Azure File Share will be mounted on Batch AI GPU clusters
- The recipe uses batch_image_label.py script to perform Distributed Batch Scoring with the given model and image datasets. The input images for evaluation will be partitioned by the MPI rank, so that each MPI worker will evaluate part of the whole image set independently.
- The expected scoring output should be text files with the following content:
ILSVRC2012_val_00000102.JPEG: Rhodesian ridgeback
ILSVRC2012_val_00000103.JPEG: tripod
ILSVRC2012_val_00000104.JPEG: typewriter keyboard
ILSVRC2012_val_00000105.JPEG: silky terrier
ILSVRC2012_val_00000106.JPEG: Windsor tie
ILSVRC2012_val_00000107.JPEG: harvestman
ILSVRC2012_val_00000108.JPEG: violin
ILSVRC2012_val_00000109.JPEG: loudspeaker
ILSVRC2012_val_00000110.JPEG: apron
ILSVRC2012_val_00000111.JPEG: American lobster
...You can find Jupyter Notebook for this recipe in TensorFlow-GPU-Distributed-Inference.ipynb.
You can find Azure CLI 2.0 instructions for this recipe in cli-instructions.md.
Under construction...
If you have any problems or questions, you can reach the Batch AI team at [email protected] or you can create an issue on GitHub.
We also welcome your contributions of additional sample notebooks, scripts, or other examples of working with Batch AI.