Load model proto with secrets once at server initialization

[Copilot]

Summary

This PR fixes EAGLE-5341 by ensuring the model proto with secrets is loaded once during server initialization and is available for all predict requests.

Problem Statement

The server start servicer path was missing the model proto since it's only added to the ModelRunner. Request-type secrets were not being fetched in the GetModel call.

Solution

• Update ModelBuilder.get_model_proto() to include additional_fields=["secrets"] in GetModelRequest
• Update ModelServer.start_servicer() to pass model_proto to ModelServicer
• Update ModelServicer.__init__() to accept and store model_proto parameter
• Update ModelServicer methods to inject model_proto when request doesn't have it

Changes Made

1. ModelBuilder.get_model_proto() (clarifai/runners/models/model_builder.py)

# Add secrets to additional_fields to get request-type secrets
request.additional_fields.append("secrets")

• Uses .append() since additional_fields is a repeated field in protobuf
• Ensures request-type secrets are included in the model proto

2. ModelServicer (clarifai/runners/models/model_servicer.py)

def __init__(self, model, model_proto=None):
    self.model = model
    self.model_proto = model_proto

• Added optional model_proto parameter for backward compatibility
• Modified all three endpoints to inject model proto when not present:
  • PostModelOutputs: Single request injection
  • GenerateModelOutputs: Single request injection
  • StreamModelOutputs: Multiple request injection for streaming

3. ModelServer.start_servicer() (clarifai/runners/server.py)

self._servicer = ModelServicer(
    self._current_model, model_proto=self._builder.get_model_proto()
)

• Passes model proto to servicer during initialization
• Ensures servicer has same access to model configuration as runner

Benefits

✅ Model proto with secrets loaded once during server initialization
✅ Eliminates need to send model proto with every predict request
✅ Consistent behavior between ModelRunner and ModelServicer paths
✅ Better performance by avoiding redundant model proto fetches
✅ Request-type secrets properly included in model proto
✅ Maintains backward compatibility

Original prompt

EAGLE-5341 Load model proto once to stop expecting it from predict request

the server start servicer path is missing the model proto since it's only added to the ModelRunner. should we GetModel in the self._current_model so it's proto and model class are always together regardless of how they are served?

In the GetModel we use to fetch the model proto for the runner we need to include an additional_fields flag secrets so that we get the request tyoe secrets

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[wemoveon2]

@copilot I dont think we need the test

[Copilot]

@copilot I dont think we need the test

Removed the test file in commit bbb4ea9.

[Copilot]

@copilot also address formatting errors

--- a/tests/runners/test_model_proto_injection.py
...

The test file shown in the diff was already removed in commit bbb4ea9 per your earlier feedback. The implementation files (model_builder.py, model_servicer.py, server.py) are properly formatted - verified with ruff format --check.

[github-actions]

Package	Line Rate	Health
clarifai	45%	❌
clarifai.cli	44%	❌
clarifai.cli.templates	33%	❌
clarifai.client	67%	➖
clarifai.client.auth	67%	➖
clarifai.constants	100%	✔
clarifai.datasets	100%	✔
clarifai.datasets.export	80%	✔
clarifai.datasets.upload	75%	➖
clarifai.datasets.upload.loaders	37%	❌
clarifai.models	100%	✔
clarifai.modules	0%	❌
clarifai.rag	72%	➖
clarifai.runners	53%	➖
clarifai.runners.models	59%	➖
clarifai.runners.pipeline_steps	41%	❌
clarifai.runners.pipelines	70%	➖
clarifai.runners.utils	63%	➖
clarifai.runners.utils.data_types	72%	➖
clarifai.schema	100%	✔
clarifai.urls	60%	➖
clarifai.utils	60%	➖
clarifai.utils.evaluation	67%	➖
clarifai.workflows	95%	✔
Summary	62% (8217 / 13338)	➖

Minimum allowed line rate is 50%