Runner Trainer

[phi-jkim]

What does this PR do?

Trainable Runner

• GradComponent Integration: Runner now inherits from GradComponent
• New forward() method for optimization (runner.py:792-950)
• Chain Predecessors: Each step's output becomes a trainable predecessor for the next step, enabling gradient flow
• Parameter Wrapping: Final results are wrapped in OutputParameter with proper gradient functions configured
• Backward Context: Automatic setup of BackwardContext for gradient computation with prompt templates and backward engines
• commented out CombineStepHistoryAndRunnerResult which follows the original ReAcT agent for optimization

Runner Trainer

• New RunnerTrainer class provides a generic interface for training Runner models (runner_trainer.py:36-100)
• comparison between new Runner workflow and original ReActAgent training

Before submitting

• Was this discussed/agreed via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you list all the breaking changes introduced by this pull request?

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB