Implementation:Arize ai Phoenix Legacy VertexAIModel

Overview

VertexAIModel is a legacy, deprecated model wrapper in the phoenix-evals package that provides an interface for using Google VertexAI text generation and code generation models (e.g., text-bison, code-bison). It extends BaseModel and integrates with the VertexAI TextGenerationModel and CodeGenerationModel classes from the google-cloud-aiplatform package. This wrapper does not support async execution and does not include rate limiting. For Gemini models, use GeminiModel or GoogleGenAIModel instead.

LLM_Evaluation Model_Integration

Description

The VertexAIModel class is implemented as a Python dataclass that extends the abstract BaseModel. Key characteristics include:

• Legacy/Deprecated status: This wrapper targets older VertexAI text and code generation models. For Gemini models, the GeminiModel or GoogleGenAIModel wrappers are recommended.
• Dual model support: Automatically selects between TextGenerationModel and CodeGenerationModel based on whether "code" appears in the model name (via the is_codey_model property).
• Tuned model support: Can load tuned/fine-tuned models via the tuned_model parameter, using get_tuned_model() instead of from_pretrained().
• No async support: The _async_generate_with_extra() method simply delegates to the synchronous _generate_with_extra(), as the underlying SDK lacks async capabilities.
• No rate limiting: Uses the default RateLimiter from the base class without configuring it with a specific error type.
• Codey-aware parameters: When using Codey (code generation) models, top_k and top_p parameters are excluded from invocation parameters since they are not supported.
• VertexAI initialization: Calls vertexai.init() with project, location, and credentials during setup.
• Deprecated field migration: Both model_name and tuned_model_name fields are deprecated with automatic migration and deprecation warnings.
• No usage tracking: Returns an empty ExtraInfo() since the legacy VertexAI predict() API does not provide token usage information.

Usage

# Set up your GCP environment
# https://cloud.google.com/vertex-ai/generative-ai/docs/start/quickstarts/quickstart-multimodal

from phoenix.evals.models import VertexAIModel

# Basic usage with defaults (text-bison)
model = VertexAIModel()

# Specify project and model
model = VertexAIModel(
    model="text-bison",
    project="my-gcp-project",
    location="us-central1",
)

response = model("Write a Python function to sort a list.")
print(response)

Code Reference

Source Location

Property	Value
Repository	Arize-ai/phoenix
File	packages/phoenix-evals/src/phoenix/evals/legacy/models/vertexai.py
Lines	215
Module	phoenix.evals.legacy.models.vertexai

Class Signature

@dataclass
class VertexAIModel(BaseModel):
    project: Optional[str] = None
    location: Optional[str] = None
    credentials: Optional["Credentials"] = None
    model: str = "text-bison"
    tuned_model: Optional[str] = None
    temperature: float = 0.0
    max_tokens: int = 1024
    top_p: float = 0.95
    top_k: int = 40
    # Deprecated
    model_name: Optional[str] = None
    tuned_model_name: Optional[str] = None

Constructor Parameters

Parameter	Type	Default	Description
project	Optional[str]	None	GCP project ID.
location	Optional[str]	None	GCP location/region (defaults to us-central1).
credentials	Optional[Credentials]	None	Google auth credentials.
model	str	"text-bison"	The VertexAI model name to use.
tuned_model	Optional[str]	None	Tuned/fine-tuned model name; if set, overrides model.
temperature	float	0.0	Sampling temperature.
max_tokens	int	1024	Maximum output tokens.
top_p	float	0.95	Nucleus sampling probability mass (ignored for Codey models).
top_k	int	40	Top-K sampling cutoff (ignored for Codey models).
model_name	Optional[str]	None	Deprecated. Use model instead.
tuned_model_name	Optional[str]	None	Deprecated. Use tuned_model instead.

Key Methods

Method	Signature	Description
__post_init__	(self) -> None	Migrates deprecated fields, initializes environment, VertexAI, and model.
_migrate_model_name	(self) -> None	Handles migration of model_name and tuned_model_name with warnings.
_init_environment	(self) -> None	Imports vertexai and google.api_core.exceptions.
_init_vertex_ai	(self) -> None	Calls vertexai.init() with project, location, and credentials.
_instantiate_model	(self) -> None	Creates TextGenerationModel or CodeGenerationModel and loads pretrained or tuned weights.
_generate_with_extra	(self, prompt, **kwargs) -> Tuple[str, ExtraInfo]	Synchronous generation via model.predict(); returns empty ExtraInfo.
_async_generate_with_extra	async (self, prompt, **kwargs) -> Tuple[str, ExtraInfo]	Delegates to synchronous generation (no native async).
is_codey_model	@property -> bool	Returns True if model name contains "code".
invocation_params	@property -> Dict[str, Any]	Returns invocation parameters, excluding top_k/top_p for Codey models.
verbose_generation_info	(self) -> str	Returns formatted invocation parameters string.

Helper Function

Function	Signature	Description
is_codey_model	(model_name: str) -> bool	Returns True if "code" is in the model name string.

Import

from phoenix.evals.models import VertexAIModel

I/O Contract

Direction	Type	Description
Input	Union[str, MultimodalPrompt]	A text string or multimodal prompt (converted to text-only).
Output	str	Generated text response.
Output (with extra)	Tuple[str, ExtraInfo]	Generated text paired with empty ExtraInfo() (no usage data available).
Error	ImportError	Raised if google-cloud-aiplatform package is not installed.

Usage Examples

Text Generation

from phoenix.evals.models import VertexAIModel

model = VertexAIModel(
    model="text-bison",
    project="my-project",
    temperature=0.2,
    max_tokens=512,
)
response = model("Explain the MapReduce programming model.")
print(response)

Code Generation

from phoenix.evals.models import VertexAIModel

model = VertexAIModel(
    model="code-bison",
    project="my-project",
    temperature=0.0,
)
# top_k and top_p are automatically excluded for Codey models
response = model("Write a Python function to compute the Fibonacci sequence.")
print(response)

With Tuned Model

from phoenix.evals.models import VertexAIModel

model = VertexAIModel(
    tuned_model="projects/my-project/locations/us-central1/models/my-tuned-model",
    project="my-project",
)
response = model("Classify the following text: ...")
print(response)

Related Pages

• Principle:Arize_ai_Phoenix_LLM_Provider_Configuration
• Heuristic:Arize_ai_Phoenix_Warning_Deprecated_VertexAIModel
• Arize_ai_Phoenix_Legacy_BaseModel - Abstract base class that VertexAIModel extends
• Arize_ai_Phoenix_Legacy_GeminiModel - Recommended wrapper for Gemini models via VertexAI SDK
• Arize_ai_Phoenix_Legacy_GoogleGenAIModel - Alternative Gemini wrapper using the Google GenAI SDK
• Arize_ai_Phoenix_Legacy_LiteLLMModel - LiteLLM wrapper that also supports VertexAI models