feat(llm): add unary chat model publisher

[furionw]

Why

Application-owned unary chat handlers currently call both register_model and serve_unary_endpoint to become public chat models. An application that calls a remote generator also has to create its endpoint client and wait for a routable worker before advertising its public model.

What changes

• Add UnaryChatModel(model_path, service_name, public_model_name=None, chat_template=None) to dynamo.llm. Its serve(runtime, handler) creates <service_name>.app.generate, registers an aggregated token-input Chat model, and serves the application handler.
• Add await LLMUnaryClient.connect(runtime, endpoint_name), which creates the endpoint client and waits for a routable instance before returning the existing unary client wrapper. The direct constructor remains available.
• The public model name defaults to the service name and can be overridden. Request logic remains in the application handler.

Full PR sequence

1. #13293 — encoder handoff (merged)
2. #15195 — unary endpoint helpers (merged)
3. #15552 — model and client utilities
4. #14859 — application-owned orchestrator example
5. #15200 — draft end-to-end verification

The three open PRs form GitHub stack #15572. The first two remain in completed stack #15197.

Verification

• Exact stacked test head 7655a83052a9c721571a6a101715337b4cbc329d (test(examples): cover the remote custom encoder topology end to end #15200), rebuilt in release mode on dl-a100 with two A100 80GB GPUs: 29 focused utility/orchestrator tests and both plain-text and image serve cases passed.
• The image case confirmed 42, the classifier label, and the logged encoder-to-generator handoff. Pre-commit and pre-merge checks passed on all three stacked PR heads.

Summary by CodeRabbit

• New Features
  • Added support for serving application-owned handlers as chat models, with configurable model names and chat templates.
  • Added an asynchronous client connection method that waits for a service instance to become available before connecting.

[devin-ai-integration]

✅ Dynamo PR CI passed — run 37032048466 (attempt 1) on 2e1226b792

Gate checks: ✅ backend-status-check · ✅ deploy-status-check · ✅ dynamo-status-check

_{Posted automatically by Devin for run 37032048466. Updated on every full-CI run of this PR.}

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository: ai-dynamo/dynamo/.coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: bec39a6d-74ae-4bf6-b03d-c7d3cbeefb9e

📥 Commits

Reviewing files that changed from the base of the PR and between e07d871 and 2e1226b.

📒 Files selected for processing (3)

• lib/bindings/python/src/dynamo/llm/__init__.py
• lib/bindings/python/src/dynamo/llm/_unary.py
• lib/bindings/python/tests/test_llm_unary.py

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 11 remain after this review.

---

Walkthrough

The Python LLM package now exports UnaryChatModel. The new model registers an application handler as an aggregated chat model and serves it through a unary endpoint. LLMUnaryClient.connect creates a client for a named runtime endpoint and waits for instances. Tests cover registration, request forwarding, readiness, and errors.

Changes

Unary chat model

Layer / File(s)	Summary
Register and serve unary chat handlers lib/bindings/python/src/dynamo/llm/__init__.py, lib/bindings/python/src/dynamo/llm/_unary.py, lib/bindings/python/tests/test_llm_unary.py	UnaryChatModel registers the configured endpoint as an aggregated chat model and serves the handler. The package exports the class. Tests cover default and explicit public names, optional templates, endpoint selection, and request forwarding.
Connect unary clients to runtime endpoints lib/bindings/python/src/dynamo/llm/_unary.py, lib/bindings/python/tests/test_llm_unary.py	LLMUnaryClient.connect obtains the named endpoint client, waits for available instances, and returns a wrapped client. Tests cover readiness and error propagation.

Priority: ➖ Normal

Estimated code review effort: 2 (Simple) | ~12 minutes

Merge Risk: ⚪ Minimal · up to 2e122

The unary chat model and client connection changes appear ready to merge after normal checks.

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 20.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 15 functions across 3 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.
Description check	⚠️ Warning	The description explains the motivation, implementation, PR sequence, and verification results. However, it does not include the required Related Issues section or identify where the reviewer should s…	Add the required Related Issues section and either link the relevant issue or check the no-related-issue confirmation. Also add a Where should the reviewer start section that names the key files or changes for review.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly identifies the main change: adding a unary chat model publisher to the LLM package.

Full details: Description check

Explanation

The description explains the motivation, implementation, PR sequence, and verification results. However, it does not include the required Related Issues section or identify where the reviewer should start.

• Fix all pre-merge checks with AI

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Autopilot is currently an internal CodeRabbit preview.

---

_{Comment @coderabbitai help to get the list of available commands.}

[jthomson04]

No actionable findings from the source review at 2e1226b. Tests and CI were not independently validated.