add fallback_to_policy_gradient option

[binary-husky]

Description

allow ignoring vllm logprob in pure on-policy mode

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 @binary-husky, 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 introduces a new configuration option within the PPO policy loss function, enabling a flexible adjustment to how gradients are computed. By providing a fallback_to_policy_gradient flag, the system can be configured to operate in a pure on-policy mode, which is useful for specific training scenarios where the traditional old_logprob comparison is not desired. This enhances the adaptability of the PPO algorithm for various reinforcement learning setups.

Highlights

• New Configuration Option: A new boolean option, fallback_to_policy_gradient, has been added to the PPOPolicyLoss class, defaulting to False.
• Policy Gradient Mode: When fallback_to_policy_gradient is enabled, the calculation of negative_approx_kl is modified from logprob - old_logprob to logprob - logprob.detach(), effectively allowing a pure on-policy gradient mode by ignoring the old_logprob.
• Debugging Output: Temporary print statements have been added within the __call__ method to display the shape of logprob when the fallback_to_policy_gradient option is active.

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 by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

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 pull request 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. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

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 adds a fallback_to_policy_gradient option to the PPO loss function. The implementation correctly falls back to a policy gradient update by recalculating negative_approx_kl when the flag is enabled. The logic is sound. My review includes two main points: first, some debugging print statements should be removed. Second, unit tests should be added to cover this new functionality and prevent future regressions. I have provided specific comments for these points.

[pan-x-c]

/unittest-module-algorithm

[github-actions]

Summary

Tests 📝	Passed ✅	Failed ❌	Skipped ⏭️	Other ❓	Flaky 🍂	Duration ⏱️
27	27	0	0	0	0	18.0s

Tests

Test Name	Status	Flaky	Duration
tests/algorithm/advantage_fn_test.py::TestGroupedAdvantageFn::test_batch_level_std_grpo	✅		41ms
tests/algorithm/advantage_fn_test.py::TestGroupedAdvantageFn::test_batch_level_step_wise_grpo_advantage	✅		2ms
tests/algorithm/advantage_fn_test.py::TestGroupedAdvantageFn::test_duplicate_grpo	✅		5ms
tests/algorithm/advantage_fn_test.py::TestGroupedAdvantageFn::test_grpo_advantage	✅		3ms
tests/algorithm/advantage_fn_test.py::TestGroupedAdvantageFn::test_grpo_correct_bias	✅		2ms
tests/algorithm/advantage_fn_test.py::TestGroupedAdvantageFn::test_grpo_reward_std	✅		1ms
tests/algorithm/advantage_fn_test.py::TestGroupedAdvantageFn::test_step_wise_grpo_advantage	✅		2ms
tests/algorithm/advantage_fn_test.py::TestGroupedAdvantageFn::test_step_wise_grpo_with_std_threshold	✅		2ms
tests/algorithm/kl_fn_test.py::KLFnTest::test_abs_kl_fn	✅		1ms
tests/algorithm/kl_fn_test.py::KLFnTest::test_corrected_k3_fallback	✅		1ms
tests/algorithm/kl_fn_test.py::KLFnTest::test_corrected_k3_loss	✅		1ms
tests/algorithm/kl_fn_test.py::KLFnTest::test_corrected_k3_same_policy	✅		1ms
tests/algorithm/kl_fn_test.py::KLFnTest::test_corrected_k3_with_old_logprob	✅		1ms
tests/algorithm/kl_fn_test.py::KLFnTest::test_dummy_kl_fn	✅		1ms
tests/algorithm/kl_fn_test.py::KLFnTest::test_k1_kl_fn	✅		1ms
tests/algorithm/kl_fn_test.py::KLFnTest::test_k2_kl_fn	✅		1ms
tests/algorithm/kl_fn_test.py::KLFnTest::test_k3_kl_fn	✅		1ms
tests/algorithm/kl_fn_test.py::KLFnTest::test_kl_loss_aggregation_modes	✅		1ms
tests/algorithm/kl_fn_test.py::KLFnTest::test_low_var_kl_fn	✅		1ms
tests/algorithm/policy_loss_test.py::VerlPolicyLossTest::test_dpo_policy_loss	✅		1ms
tests/algorithm/policy_loss_test.py::VerlPolicyLossTest::test_gspo_policy_loss	✅		1ms
tests/algorithm/policy_loss_test.py::VerlPolicyLossTest::test_mix_policy_loss	✅		1ms
tests/algorithm/policy_loss_test.py::VerlPolicyLossTest::test_opmd_policy_loss	✅		1ms
tests/algorithm/policy_loss_test.py::VerlPolicyLossTest::test_ppo_policy_loss	✅		1ms
tests/algorithm/policy_loss_test.py::VerlPolicyLossTest::test_ppo_policy_loss_with_sequence_masking	✅		1ms
tests/algorithm/policy_loss_test.py::VerlPolicyLossTest::test_sapo_policy_loss	✅		1ms
tests/algorithm/policy_loss_test.py::VerlPolicyLossTest::test_sft_policy_loss	✅		1ms

Github Test Reporter by CTRF 💚

[yanxi-chen]

lgtm