Heuristic:Arize ai Phoenix Warning Deprecated VertexAIModel
| Knowledge Sources | |
|---|---|
| Domains | LLM_Providers, Deprecation |
| Last Updated | 2026-02-14 06:00 GMT |
Overview
Deprecation warning: VertexAIModel (legacy) has deprecated fields and resides in the legacy module. Use GeminiModel or the modern v2 adapter system instead.
Description
The VertexAIModel class in phoenix.evals.legacy.models.vertexai has two deprecated fields that issue runtime DeprecationWarning messages:
- model_name is deprecated since v3.0.0 — use model instead
- tuned_model_name is deprecated since v3.0.0 — use tuned_model instead
Additionally, VertexAIModel resides in the legacy evals module. The modern v2 API uses the adapter pattern (GoogleGenAIAdapter in phoenix.evals.llm.adapters.google) which supports Vertex AI through the Google GenAI SDK.
Usage
This warning applies when using VertexAIModel from the legacy evals module. Consider migrating to the v2 adapter-based LLM class with a Google GenAI adapter for Vertex AI workloads.
The Insight (Rule of Thumb)
- Action: Replace model_name parameter with model and tuned_model_name with tuned_model
- Value: Avoids runtime deprecation warnings; prepares for removal in future release
- Trade-off: Migration to v2 adapter system requires restructuring LLM initialization code
Reasoning
The legacy VertexAIModel wraps the vertexai SDK directly, while the modern adapter system wraps google.generativeai which provides a superset of functionality including both Gemini and Vertex AI access. The field renames align with standard naming conventions across all provider models.