Add std_thresholdoption to StepWiseGRPOAdvantageFn, to filter out zero-grad group samples.

[garyzhang99]

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 @garyzhang99, 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 enhances the StepWiseGRPOAdvantageFn by adding a std_threshold parameter. This addition enables the filtering of task groups with low reward standard deviation, which can improve the stability and efficiency of the learning process. The changes include modifications to the advantage function logic, metric tracking, and the addition of a new unit test to validate the new functionality.

Highlights

• New Feature: std_threshold: Introduces a std_threshold option to the StepWiseGRPOAdvantageFn. This allows filtering out zero-grad group samples based on the standard deviation of rewards within a task group.
• Filtering Mechanism: Task groups with a reward standard deviation equal to or below the std_threshold are now skipped during advantage calculation.
• Metrics Tracking: The code now tracks and reports the number of experiences filtered out due to the std_threshold in the filtered_count metric.
• Unit Test: A new unit test, test_step_wise_grpo_reward_std, has been added to verify the filtering mechanism when the reward standard deviation is zero.

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 std_threshold option to StepWiseGRPOAdvantageFn to filter out experience groups with a reward standard deviation below a given value. This is useful for avoiding issues with zero-gradient samples. The implementation is sound and includes a corresponding unit test. My review includes one suggestion to correct the filtering logic to avoid unintentionally filtering out groups that contain only a single experience.

[garyzhang99]

/unittest-module-algorithm

[github-actions]

Summary

Tests 📝	Passed ✅	Failed ❌	Skipped ⏭️	Other ❓	Flaky 🍂	Duration ⏱️
14	14	0	0	0	0	10.8s

Tests

Test Name	Status	Flaky	Duration
tests/algorithm/advantage_fn_test.py::TestGroupedAdvantageFn::test_batch_level_std_grpo	✅		40ms
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/policy_loss_test.py::VerlPolicyLossTest::test_dpo_policy_loss	✅		2ms
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_sft_policy_loss	✅		1ms

Github Test Reporter by CTRF 💚

[garyzhang99]

/unittest-module-algorithm

[github-actions]

Summary

Tests 📝	Passed ✅	Failed ❌	Skipped ⏭️	Other ❓	Flaky 🍂	Duration ⏱️
14	14	0	0	0	0	11.3s

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/policy_loss_test.py::VerlPolicyLossTest::test_dpo_policy_loss	✅		2ms
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_sft_policy_loss	✅		1ms

Github Test Reporter by CTRF 💚

[hiyuchang]

/unittest-module-common

[github-actions]

Summary

Tests 📝	Passed ✅	Failed ❌	Skipped ⏭️	Other ❓	Flaky 🍂	Duration ⏱️
33	33	0	0	0	0	4m 56s

Tests

Test Name	Status	Flaky	Duration
tests/common/config_test.py::TestConfig::test_all_examples_are_valid	✅		40.9s
tests/common/config_test.py::TestConfig::test_config_flatten	✅		38ms
tests/common/config_test.py::TestConfig::test_continue_from_checkpoint_is_valid	✅		185ms
tests/common/config_test.py::TestConfig::test_default_workflow	✅		87ms
tests/common/config_test.py::TestConfig::test_load_default_config	✅		4.0s
tests/common/config_test.py::TestConfig::test_max_token_len_per_gpu_set_correctly	✅		90ms
tests/common/config_test.py::TestConfig::test_update_config_from_ray_cluster	✅		1.8s
tests/common/experience_test.py::TestEID::test_eid_properties	✅		1ms
tests/common/experience_test.py::TestExperience::test_action_mask_and_logprobs_type	✅		1ms
tests/common/experience_test.py::TestExperience::test_assertions	✅		1ms
tests/common/experience_test.py::TestExperience::test_dpo_experience	✅		1ms
tests/common/experience_test.py::TestExperience::test_gather	✅		1ms
tests/common/experience_test.py::TestExperience::test_hf_datasets_conversion	✅		18ms
tests/common/experience_test.py::TestExperience::test_multi_turn_experience	✅		1ms
tests/common/experience_test.py::TestExperience::test_serialize_deserialize	✅		2ms
tests/common/experience_test.py::TestExperience::test_single_turn_experience	✅		1ms
tests/common/experience_test.py::TestExperience::test_to_dict	✅		1ms
tests/common/experience_test.py::TestExperienceConversion::test_batch_conversion	✅		1ms
tests/common/experience_test.py::TestExperienceConversion::test_dpo_experience_batch_conversion	✅		1ms
tests/common/experience_test.py::TestExperienceConversion::test_experience_model_experience_conversion	✅		1ms
tests/common/experience_test.py::TestExperienceConversion::test_gather_experiences_with_custom_fields	✅		1ms
tests/common/experience_test.py::TestExperienceConversion::test_multiturn_experience_batch_converstion	✅		1ms
tests/common/vllm_test.py::ModelWrapperTest_0::test_generate	✅		50.7s
tests/common/vllm_test.py::ModelWrapperTest_1::test_generate	✅		32.4s
tests/common/vllm_test.py::ModelWrapperTest_2::test_generate	✅		43.8s
tests/common/vllm_test.py::TestModelLen_0::test_model_len	✅		17.2s
tests/common/vllm_test.py::TestModelLen_1::test_model_len	✅		17.1s
tests/common/vllm_test.py::TestAPIServer::test_api	✅		23.2s
tests/common/vllm_test.py::TestAsyncAPIServer::test_api_async	✅		22.7s
tests/common/vllm_test.py::TestTokenizer::test_action_mask	✅		258ms
tests/common/vllm_test.py::TestTokenizer::test_action_mask_with_tools	✅		250ms
tests/common/vllm_test.py::TestAPIServerToolCall_0_deepseek_r1::test_api_tool_calls	✅		19.7s
tests/common/vllm_test.py::TestAPIServerToolCall_1::test_api_tool_calls	✅		17.8s

Github Test Reporter by CTRF 💚