### What happened?

I am attempting to use a custom proxy for Vertex AI models by setting a custom api_base. My proxy does not require Google Vertex credentials.

However, the library crashes with DefaultCredentialsError.

I noticed that litellm/llms/vertex_ai/vertex_gemini_transformation.py ([ref](https://github.com/BerriAI/litellm/commit/5cfcc98d2ea7c0bf834f60d699e9b4fbf49a5010#diff-e87dcc905ac5cf41f7df6c10acaf4b673ed26ef40c911b93aa6c6fae6e6afa43R159-R161)) contains logic to skip credential validation when api_base is present. However, my execution flow does not reach that transformation class. Instead, it hits vertex_llm_base.py, which unconditionally attempts to load authentication, ignoring the presence of a custom api_base.

### Steps to Reproduce

1. Call litellm.completion with a Vertex model (e.g., vertex_ai/chat-bison or similar). 2. Pass a custom api_base (e.g., a local proxy). 3. Ensure no Google Cloud credentials are set in the environment. 4. Run the code.

### Relevant log output

```shell Traceback (most recent call last): File "/venv/lib/python3.11/site-packages/litellm/llms/vertex_ai/vertex_llm_base.py", line 488, in get_access_token _credentials, credential_proj…