| Error |
Cause |
Fix |
| No inference API is accessible for model |
Model access not enabled or IAM permission missing |
Enable model access in Bedrock console; verify IAM policy includes bedrock:InvokeModelWithResponseStream |
| Input file is empty |
S3 object has 0 bytes |
Verify the input JSONL file has content |
| File is not valid JSONL |
Malformed JSON in input file |
Validate each line is valid JSON |
| AccessDenied on S3 |
IAM role lacks S3 permissions |
Add s3:GetObject for input path and s3:PutObject for output path |
| ValidationException: multiple evaluation methods |
Provided both steeringCriteria AND customLLMJConfig or Lambda |
Use only ONE evaluation method per template |
| ValidationException: missing label |
Missing customEvaluationMetricLabel when using LLMJ or Lambda |
Add the customEvaluationMetricLabel field |
| Silent failure: inputVariables |
Multiple keys in one inputVariables object |
Use one key per object in the array |
| ValidationException: placeholder mismatch |
inputVariables keys don't match {{placeholders}} in template |
Ensure keys exactly match placeholder names |
| ValidationException: wrong bracket syntax |
Used {variable} instead of {{variable}} |
Use double curly brackets |
compressionRatio is 1.0 and the original prompt was returned |
With shortenPrompt enabled, the optimized prompt was not both shorter than the original and at least as high scoring. This also occurs when the original prompt already scores at the maximum on your evaluation samples, or when the samples do not distinguish between candidate prompts, because the optimizer compares candidates by evaluation score. |
Include evaluation samples that the original prompt does not fully pass but that a better prompt could pass, then run the job again. A prompt that is already concise may not compress further. |
| A newly launched model is not available for selection, or a job fails for a specific model |
The model has not yet been enabled for Advanced Prompt Optimization, or it is not available in the job's Region |
Confirm the model is available in your Region. If it is, try again later. If the model is still unavailable, contact AWS Support. |
| ValidationException: application inference profile could not be found, or resolves to a model that is not an approved judge model |
The customLLMJModelId application inference profile does not exist in your account, or it routes to a model that is not a supported judge model |
Use an application inference profile in the same account that routes to a model on the Supported judge models list |