feat: add Nemotron model support with message-based prompts

[Pouyanpi]

Description

This PR adds support for Nemotron models in NeMo Guardrails with a message-based prompt approach. It implements a system for controlling detailed thinking traces both for internal tasks and in final user responses, similar to how it works on build.nvidia.com.

Key Changes

• Added new nemotron.yml with message-based prompt templates
• Implemented "detailed thinking on" system messages for complex reasoning tasks
• Added system message handling in LLMRails to control thinking traces
• Provided example configuration in examples/configs/nemotron

Features

Control Detailed Thinking via System Messages

As Nemotron is a hybrid reasoning model, users can toggle the "detailed thinking" feature for final responses:

# With detailed thinking enabled
response = rails.generate(messages=[
    {"role": "system", "content": "detailed thinking on"}, 
    {"role": "user", "content": "How is the weather today?"}
])

# Result includes thinking process
{'role': 'assistant',
 'content': '<think>\n</think>I\'m sorry, but I don\'t know the weather...'}

# Without detailed thinking (standard mode)
response = rails.generate(messages=[
    {"role": "user", "content": "How is the weather today?"}
])

# Result without thinking process
{'role': 'assistant',
 'content': 'The weather! Unfortunately, I don\'t have real-time access...'}

Configuration Options

Control detailed thinking for internal tasks:

models:
  - type: main
    engine: nim
    model: nvidia/llama-3.1-nemotron-ultra-253b-v1
    reasoning_config:
      remove_reasoning_traces: False # set to true to remove traces

Documentation

Added README.md with examples and explanations of the prompt system and thinking control options.

Testing

Added tests to verify:

• Nemotron models use message-based prompts
• Tasks like generate_bot_message have "detailed thinking on"
• Tasks like generate_user_intent don't have "detailed thinking on"

[github-actions]

Documentation preview

https://nvidia.github.io/NeMo-Guardrails/review/pr-1199

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 91.66667% with 1 line in your changes missing coverage. Please review.

Project coverage is 68.65%. Comparing base (36d625e) to head (2b37d04).
Report is 3 commits behind head on develop.

Files with missing lines	Patch %	Lines
nemoguardrails/llm/prompts.py	87.50%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1199      +/-   ##
===========================================
+ Coverage    68.43%   68.65%   +0.21%     
===========================================
  Files          161      161              
  Lines        15943    15978      +35     
===========================================
+ Hits         10910    10969      +59     
+ Misses        5033     5009      -24     

Flag	Coverage Δ
python	68.65% <91.66%> (+0.21%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
nemoguardrails/rails/llm/llmrails.py	87.21% <100.00%> (+0.09%)	⬆️
nemoguardrails/llm/prompts.py	90.47% <87.50%> (-0.32%)	⬇️

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[Copilot]

Pull Request Overview

This PR adds support for Nemotron models in NeMo Guardrails using message‐based prompt templates with a detailed thinking feature. Key changes include adding new YAML prompt definitions for Nemotron and DeepSeek, implementing system message conversion in LLMRails, and updating test suites and configuration files accordingly.

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
tests/test_system_message_conversion.py	Added tests to ensure system messages are converted correctly.
tests/test_nemotron_prompt_modes.py	Added tests verifying detailed thinking behavior for Nemotron prompts.
nemoguardrails/rails/llm/llmrails.py	Modified to convert system messages to SystemMessage events appropriately.
nemoguardrails/llm/prompts/nemotron_reasoning.yml	New prompt definitions with detailed thinking enabled in specific tasks.
nemoguardrails/llm/prompts/llama3.yml	Updated prompt definitions to include new model identifiers.
nemoguardrails/llm/prompts.py	Updated scoring logic in prompt selection for improved model matching.
examples/configs/nemotron/config.yml	New configuration file for Nemotron model support.
examples/configs/nemotron/README.md	Added documentation for using Nemotron message-based prompts.

Comments suppressed due to low confidence (1)

nemoguardrails/llm/prompts.py:108

• [nitpick] The added 'break' after detecting a substring match may prevent subsequent models in the list from being evaluated. Consider reviewing the order of model checks or accumulating all candidate scores to ensure the best matching prompt is selected.

elif _model in model:

[mikemckiernan]

lgtm, thanks so much!

[Pouyanpi]

LGTM!