Fix alert triage agent: switch to nemotron-3-nano model and improve prompts

[hsin-c]

Summary

• Switch LLM model from nvidia/llama-3.3-nemotron-super-49b-v1.5 to nvidia/nemotron-3-nano-30b-a3b across all LLM configs in config_offline_mode.yml, with temperature: 0 and max_tokens: 16384 for deterministic and more complete outputs.
• Improve triage agent prompts to produce more accurate diagnoses, especially for broad alerts like InstanceDown where the root cause could span hardware, network, or software — the agent is now instructed to use all available tools before drawing conclusions.
• Clarify categorizer prompt definitions for false_positive and need_investigation categories to reduce misclassification — false_positive now requires all diagnostic checks to indicate a healthy system, and need_investigation is reserved for genuinely incomplete or conflicting data.
• Fix warning in categorizer.py: replace result.text() (method call) with result.text (property access).
• Fix typo: "CPR" → "CPU" in host_performance_check tool description.

Test plan

• Ran pytest -v --run_slow --run_integration examples/advanced_agents/alert_triage_agent/tests/test_alert_triage_agent_workflow.py::test_full_workflow 10 times — all 10 runs passed with no intermittent failures.

Summary by CodeRabbit

• Bug Fixes

  • Enhanced alert categorization guidance with clearer instructions on diagnostic tool usage and deeper root cause analysis
  • Improved categorization definitions to reduce false positives and better identify alerts requiring further investigation

• Improvements

  • Updated LLM model configurations with optimized parameters for enhanced performance

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

Walkthrough

The pull request updates an alert triage agent system by correcting a method call to property access in the categorizer logic, migrating LLM models to a different variant with adjusted hyperparameters in the offline configuration, and refining prompt guidance for improved diagnostic tool usage and classification accuracy.

Changes

Cohort / File(s)	Summary
Code Logic Fix examples/advanced_agents/alert_triage_agent/src/nat_alert_triage_agent/categorizer.py	Changed result.text() method call to result.text property access when computing report_content.
LLM Configuration Updates examples/advanced_agents/alert_triage_agent/src/nat_alert_triage_agent/configs/config_offline_mode.yml	Updated 6 LLM configurations: migrated model from nvidia/llama-3.3-nemotron-super-49b-v1.5 to nvidia/nemotron-3-nano-30b-a3b; set temperature to 0; increased max_tokens to 16384; removed top_p setting where applicable. Affects ata_agent_llm, tool_reasoning_llm, telemetry_metrics_analysis_agent_llm, maintenance_check_llm, categorizer_llm, and nim_rag_eval_llm.
Prompt Refinements examples/advanced_agents/alert_triage_agent/src/nat_alert_triage_agent/prompts.py	Expanded diagnostic tool guidance emphasizing comprehensive tool usage for broad alerts; clarified that network unreachability is a symptom not root cause; refined CategorizerPrompts definitions with more explicit, evidence-based explanations for false_positive and need_investigation categories.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately describes the main changes: switching to nemotron-3-nano model and improving prompts, with all three file modifications reflected.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

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

[mnajafian-nv]

LGTM!

[mnajafian-nv]

/merge

[willkill07]

/ok to test 27489a2

[dagardner-nv]

/ok to test 27489a2

[dagardner-nv]

/ok to test 27489a2