fix: update _evals_common to be compatible with litellm >=1.83.0

[quad2524]

Summary
This PR updates the litellm dependency to version 1.83.0 or higher. This upgrade is necessary to bypass known security vulnerabilities present in version 1.82.7.

Because litellm introduced changes to how models and providers are validated, I have also updated the internal utility functions and associated tests to maintain compatibility.

Changes
Dependency Update: Bumped litellm version requirement in setup files.

Core Logic (_evals_common): Updated _is_litellm_model to utilize the newer get_llm_provider pattern. This ensures we accurately validate model strings against LiteLLM’s supported provider list.

Test Suite:

Refactored mocks to account for the new return signature of litellm.get_llm_provider, which now includes additional metadata (model, provider, etc.).

Updated get_valid_models mocks to ensure consistent behavior during unit testing.

Fixes #6598 🦕

[matthew29tang]

Can you resolve the merge conflicts? Then I can run the workflow checks

[quad2524]

Can you resolve the merge conflicts? Then I can run the workflow checks

Done

[jsondai]

LGTM

[matthew29tang]

Hi, the lint has failed. Can you fix it?

nox > black --check --diff docs google vertexai tests noxfile.py setup.py
--- /tmpfs/src/github/python-aiplatform/vertexai/_genai/_evals_common.py	2026-04-20 22:00:03+00:00
+++ /tmpfs/src/github/python-aiplatform/vertexai/_genai/_evals_common.py	2026-04-20 22:30:57.589545+00:00
@@ -738,11 +738,11 @@
     if litellm is None:
         return False
 
     try:
         litellm.get_llm_provider(model)
-        return True 
+        return True
     except ValueError:
         return False
 
 
 def _is_gemini_model(model: str) -> bool:
would reformat /tmpfs/src/github/python-aiplatform/vertexai/_genai/_evals_common.py
--- /tmpfs/src/github/python-aiplatform/tests/unit/vertexai/genai/test_evals.py	2026-04-20 22:00:03+00:00
+++ /tmpfs/src/github/python-aiplatform/tests/unit/vertexai/genai/test_evals.py	2026-04-20 22:31:04.453888+00:00
@@ -3673,16 +3673,22 @@
     def test_run_inference_with_litellm_openai_request_format(
         self,
         mock_api_client_fixture,
     ):
         """Tests inference with LiteLLM where the row contains a chat completion request body."""
-        with mock.patch(
-            "vertexai._genai._evals_common.litellm"
-        ) as mock_litellm, mock.patch(
-            "vertexai._genai._evals_common._call_litellm_completion"
-        ) as mock_call_litellm_completion:
-            mock_litellm.get_llm_provider.return_value = ("gpt-4o", "openai", None , None)
+        with (
+            mock.patch("vertexai._genai._evals_common.litellm") as mock_litellm,
+            mock.patch(
+                "vertexai._genai._evals_common._call_litellm_completion"
+            ) as mock_call_litellm_completion,
+        ):
+            mock_litellm.get_llm_provider.return_value = (
+                "gpt-4o",
+                "openai",
+                None,
+                None,
+            )
             prompt_df = pd.DataFrame(
                 [
                     {
                         "model": "gpt-4o",
                         "messages": [
would reformat /tmpfs/src/github/python-aiplatform/tests/unit/vertexai/genai/test_evals.py

[quad2524]

Hi, the lint has failed. Can you fix it?

Fixed linting

[matthew29tang]

Appreciate the quick turnarounds on the PR feedback.

We have received some urgent internal guidance that we want to maintain the same lower bound, but we need to have a stricter upper bound on a minor version given the impact that the litellm vulnerability had.

I plan to proceed with #6617, and then we can merge in your PR with the updates to the evals files. Is that okay with you?

[quad2524]

That works for me. Thanks!

[matthew29tang]

Great, #6617 is now merged in. Can you revert your changes in setup.py?

[matthew29tang]

Thanks for the contribution! It's now merged as ac5a5e4