[EAGLE-5341] Remove model_proto caching from ModelRunner, ModelServicer, and server

[Copilot]

Why

• Model proto caching and merging logic is no longer needed in the SDK
• Requests will contain model proto directly when required, eliminating need for SDK-side caching
• Simplifies the codebase by removing unnecessary complexity

How

• Removed model_proto parameter from ModelRunner.__init__ and ModelServicer.__init__
• Removed self.model_proto instance variable from both classes
• Removed all model proto merging/copying logic from:
  • ModelRunner: runner_item_predict(), runner_item_generate(), pmo_iterator()
  • ModelServicer: PostModelOutputs(), GenerateModelOutputs(), StreamModelOutputs()
• Removed resources_pb2 import from model_servicer.py (no longer needed)
• Removed model_proto argument from ModelServicer and ModelRunner initialization in server.py
• Deleted tests/runners/test_model_proto_merge.py (tested removed functionality)

Tests

• Syntax validation passed
• Linting passed (ruff)
• Formatting verified
• Code compiles successfully

Notes

• Requests now handle model proto directly without SDK-side caching or merging
• This simplifies the code and moves responsibility to the request layer where it belongs
• No breaking changes for callers who don't use the removed model_proto parameter

Original prompt

In clarifai/runners/models/model_runner.py and clarifai/runners/models/model_servicer.py, we should merge the cached model proto with the model proto in the request instead of only doing so when the request is misisng the model proto

The request will contain the model and model version proto when the user provided request type secrets or overrides the param specs inference parameters. When it is included, it will pretty much be a minimal model and model version proto that contains the IDs and model_version.output_info which contains the inference params and the request type secret

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[wemoveon2]

@copilot why not something like

if self.model_proto is not None:
if not request.HasField("model"):
request.model.CopyFrom(self.model_proto)
else:
merged_model = resources_pb2.Model()
merged_model.CopyFrom(self.model_proto)
merged_model.MergeFrom(request.model)
request.model.CopyFrom(merged_model)

[Copilot]

@copilot why not something like

if self.model_proto is not None:
...

Good suggestion! Updated to use the explicit if-else pattern for better readability. The two code paths are now clearly separated:

• No model in request → direct CopyFrom
• Model in request → create merged_model, copy cached, merge request

Changes in commit ccea79e. All tests still passing.

[wemoveon2]

@copilot describe the behaviors tested in the tests you added

[Copilot]

@copilot describe the behaviors tested in the tests you added

The tests validate two key scenarios across different code paths:

Scenario 1: Empty request (no model proto)

• Tests verify cached model proto is directly copied to empty requests
• Validates all cached fields (id, user_id, app_id) appear in request
• Covered in: test_pmo_iterator_with_empty_request_model, test_servicer_post_model_outputs_with_empty_request

Scenario 2: Request with minimal model proto (the bug case)

• Tests verify both cached AND request fields are preserved after merge
• Request fields (id, version, output_info with inference params) override cached values
• Cached fields (user_id, app_id) that aren't in request are added
• Validates the core fix: inference params from request work alongside cached config
• Covered in: test_pmo_iterator_with_request_model_override, test_servicer_post_model_outputs_with_request_override, test_servicer_generate_model_outputs_merge, test_servicer_stream_model_outputs_merge

Each test also verifies the operation succeeds (SUCCESS status code) and covers different API methods (predict, generate, stream).

[wemoveon2]

@copilot drop passing of model proto entirely. the request will contain a model proto if it needs it, does not need anything on the sdk side.

remove the usage of the cached model proto from both the model runner and grpc servicer

[Copilot]

@copilot drop passing of model proto entirely. the request will contain a model proto if it needs it, does not need anything on the sdk side.

remove the usage of the cached model proto from both th...

Removed all model_proto caching logic as requested. Changes in commit 6730513:

• Removed model_proto parameter from ModelRunner and ModelServicer
• Removed all merging/copying logic in both runner and servicer
• Removed model_proto parameter from pmo_iterator
• Deleted test_model_proto_merge.py
• Requests now handle model proto directly

[wemoveon2]

@copilot we need to stop passing the model proto in server.py

[Copilot]

@copilot we need to stop passing the model proto in server.py

Removed model_proto from server.py. Changes in commit 5ac9566:

• Removed model_proto argument when creating ModelServicer
• Removed model_proto argument when creating ModelRunner
• Server no longer passes model proto to either component

[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	61%	➖
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% (8364 / 13480)	➖

Minimum allowed line rate is 50%