fix(bedrock): handle cohere v4 embed response dictionary format

[AndyForest]

Title

Fix parsing of bedrock cohere embed v4 responses

Relevant issues

This PR fixes an issue where bedrock/cohere.embed-v4 models return a dictionary of embeddings (keyed by type, e.g., float, int8) instead of a flat list, causing LiteLLM to incorrectly parse the response.

Previously, LiteLLM would iterate over the dictionary keys (returning ["float", "int8"]) instead of the actual embedding vectors. This change explicitly checks if the embeddings field in the response is a dictionary and handles it as embeddings_by_type.

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• I have added a screenshot of my new test passing locally
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

I have also verified this in a live integration environment using bedrock/cohere.embed-v4:0 as well as confirming other embedding models are not affected such as bedrock/cohere.embed-multilingual-v3 and bedrock/amazon.titan-embed-text-v2:0.

Type

🐛 Bug Fix
✅ Test

Changes

litellm/llms/cohere/embed/v1_transformation.py
Added new code here to check the returned data.

tests/test_litellm/llms/bedrock/embed/test_bedrock_cohere_v4.py
I have added a new unit test that mocks the Cohere v4 response structure.

[vercel]

@AndyForest is attempting to deploy a commit to the CLERKIEAI Team on Vercel.

A member of the Team first needs to authorize it.