Bug fix for Scheduler and torch.tensor

[chenyushuo]

Description

As the title says.

Checklist

Please check the following items before code is ready to be reviewed.

• [] Code has passed all tests
• Docstrings have been added/updated in Google Style
• Documentation has been updated
• Code is ready for review

[gemini-code-assist]

Summary of Changes

Hello @chenyushuo, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request focuses on addressing bug fixes and improving robustness within the Scheduler component and related torch.tensor operations. It standardizes data types for numerical computations in several algorithm components and refines the Scheduler's internal task management to ensure correct and immutable propagation of task parameters, including a new test to validate multi-step execution.

Highlights

• PyTorch Tensor Type Consistency: Explicitly set dtype=torch.float32 when creating torch.tensor objects from lists of rewards or scores across various advantage functions (add_strategy.py, grpo_advantage.py, opmd_advantage.py, rloo_advantage.py). This addresses potential type-related issues and ensures consistent float precision in numerical computations.
• Scheduler Task Handling Refinement: Refactored the Scheduler's task splitting and submission logic. The repeat_times field was removed from TaskWrapper, and instead, the rollout_args.n attribute of the Task object itself is now modified using dataclasses.replace. This ensures that the correct number of repeats is propagated directly within the task object and improves immutability.
• Flexible Task Identifiers: Updated the Task dataclass to allow batch_id and task_id to be either an int or a str (Union[int, str]), providing greater flexibility in how tasks are identified within the system.
• New Scheduler Test Case: Added a new test test_multi_step_execution in scheduler_test.py to validate the Scheduler's behavior when handling tasks across multiple steps and with max_repeat_times_per_runner configured, ensuring the refactored logic works as expected.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in issue comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments or fill out our survey to provide feedback.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces two main fixes. First, it addresses a bug in the Scheduler where task objects were mutated, causing issues with task reuse. The change to use dataclasses.replace for creating new task instances is a solid improvement for correctness and code clarity. The new test test_multi_step_execution effectively validates this fix.

Second, the PR fixes potential torch.tensor dtype inference issues by explicitly setting dtype=torch.float32. My review includes a few suggestions to apply this fix consistently across all relevant code paths and to improve the clarity of tensor creation in one of the advantage functions. I've also pointed out a leftover debug print statement in a test.

Overall, these are valuable changes. Addressing the feedback will further improve the code quality.

[chenyushuo]

/unittest-module-explorer

[github-actions]

Summary

Tests 📝	Passed ✅	Failed ❌	Skipped ⏭️	Other ❓	Flaky 🍂	Duration ⏱️
19	19	0	0	0	0	338ms

Tests

Test Name	Status	Flaky	Duration
tests/explorer/explorer_test.py::BaseExplorerCase::test_explorer	✅		2ms
tests/explorer/explorer_test.py::TestExplorerCountdownEval::test_explorer	✅		94ms
tests/explorer/explorer_test.py::TestExplorerCountdownNoEval::test_explorer	✅		92ms
tests/explorer/explorer_test.py::TestExplorerWithAddStrategy::test_explorer	✅		60ms
tests/explorer/scheduler_test.py::SchedulerTest::test_concurrent_operations	✅		4ms
tests/explorer/scheduler_test.py::SchedulerTest::test_get_results	✅		19ms
tests/explorer/scheduler_test.py::SchedulerTest::test_multi_step_execution	✅		4ms
tests/explorer/scheduler_test.py::SchedulerTest::test_scheduler_all_methods	✅		14ms
tests/explorer/scheduler_test.py::SchedulerTest::test_scheduler_restart_after_stop	✅		8ms
tests/explorer/scheduler_test.py::SchedulerTest::test_split_tasks	✅		8ms
tests/explorer/scheduler_test.py::SchedulerTest::test_wait_all	✅		7ms
tests/explorer/scheduler_test.py::SchedulerTest::test_wait_all_timeout_with_multi_batch	✅		13ms
tests/explorer/workflow_test.py::WorkflowTest::test_gsm8k_workflow	✅		1ms
tests/explorer/workflow_test.py::WorkflowTest::test_math_boxed_workflow	✅		1ms
tests/explorer/workflow_test.py::WorkflowTest::test_math_complex_workflow	✅		1ms
tests/explorer/workflow_test.py::WorkflowTest::test_math_fraction_workflow	✅		1ms
tests/explorer/workflow_test.py::WorkflowTest::test_math_workflow	✅		1ms
tests/explorer/workflow_test.py::WorkflowTest::test_rm_gallery_workflow	✅		1ms
tests/explorer/workflow_test.py::WorkflowTest::test_workflow_resettable	✅		1ms

Github Test Reporter by CTRF 💚