Troubleshooting

Ensure deployment completed successfully

Verify model deployment with cluster unit list

Check name and version in job parameters and compare it with the deployment unit name and version

Ensure you have actual model files (not .metadata or .lock files)

Check that files are readable and not Git LFS artifacts

Make sure that the deployment unit exists

Make sure that the type parameter in the JobParameters matches the engine that the model has been created on (pytorch, tensorflow, onnx)

Verify you have the correct model files for your model type:

Model weights: .onnx, or .pb, or *.pt

Model config: config.json (except for ONNX - they are self-contained models)

Model variable directory

model-directory/
├── saved_model.pb
└── variables/
    ├── variables.data-00000-of-00001
    └── variables.index

Tokenizer: tokenizer.json, tokenizer_config.json (for some models)

Vocabulary: vocab.txt, special_tokens_map.json (some NLP models)

Make sure that you have used the correct translator for the model

Make sure your custom compute job and marshaller is deployed correctly (optional: in case there is user-defined input/output/translator/translatorFactory)

Make sure that the input and output classes match the translator that has been used