Skip to content

Validation Exception when using images with llama-3.2 on Bedrock #4415

@reciprocal-space

Description

@reciprocal-space

Describe the bug

We have been using the converse operation on bedrock with us.meta.llama3-2-11b-instruct-v1:0 and image inputs but as of this past Monday the operation has begun failing.

Regression Issue

  • Select this option if this issue appears to be a regression.

Expected Behavior

We expect to receive back a successful response with the information required by our prompt.

Current Behavior

We receive the following error response:

An error occurred (ValidationException) when calling the Converse operation: The model returned the following errors: The provided image does not match the specified image format.

Reproduction Steps

# Python v3.8.12, boto 1.34.152

file_obj = s3.get_object(Bucket=bucket, Key=key)
image_content = file_obj['Body'].read()

"messages": [
    {
        "content": [
            {"text": prompt},
            {"image": {"format": "jpeg", "source": {"bytes": image_content}}}
        ],
        "role": "user"
    }
]

response = bedrock_runtime.converse(
    modelId='us.meta.llama3-2-11b-instruct-v1:0',
    messages=messages,
    inferenceConfig={
        "maxTokens": kwargs.get('max_tokens', 100),
        "temperature": kwargs.get('temperature', .5),
        "topP": kwargs.get('topP', 0.9),
    }
)

Possible Solution

No response

Additional Information/Context

No response

SDK version used

1.34.152

Environment details (OS name and version, etc.)

Docker, python:3.8.12-buster image

Metadata

Metadata

Labels

bedrockbedrock-runtimebugThis issue is a confirmed bug.p2This is a standard priority issueresponse-requestedWaiting on additional information or feedback.service-apiThis issue is caused by the service API, not the SDK implementation.

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions