SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.21492 · 2026

GradAlign: Gradient-Aligned Data Selection for LLM Reinforcement Learning

Yiming Yang, Weiwei Sun, Sean Welleck, Weihua Du, Ningyuan Yang

arXiv · PDF · Open in the Atlas

Code that ran

13 of 17 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 10 failures of our instrument; 4 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2602.21492.json · badge: /api/ran/2602.21492.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
StigLidu/GradAlign canonical (classifier label)
classifier basis: census
13 ran · 4 no recorded run
FunctionStatusWhere it lives
aggregate_similarity Ran
ran; our contract draft was wrong, not the code
StigLidu/GradAlign/automated/aggregate.py
pointer only (licence: NONE) · get_code("f8d2cea71368ec06")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

build_exp_name Ran
ran; our contract draft was wrong, not the code
StigLidu/GradAlign/automated/launch_verl_training.py
pointer only (licence: NONE) · get_code("60f309dc7281baa7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

build_local_dir Ran
ran; our contract draft was wrong, not the code
StigLidu/GradAlign/automated/merge_model.py
pointer only (licence: NONE) · get_code("b4b4346238c8b44a")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

build_target_dir Ran
ran; our contract draft was wrong, not the code
StigLidu/GradAlign/automated/merge_model.py
pointer only (licence: NONE) · get_code("5149292e77619ae5")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

compute_grpo_advantages Ran
ran; our contract draft was wrong, not the code
StigLidu/GradAlign/select/grpo_loss.py
pointer only (licence: NONE) · get_code("df7f7b3db9111a60")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

compute_kl_loss Ran
ran, honoured the contract we drafted
StigLidu/GradAlign/select/grpo_loss.py
pointer only (licence: NONE) · get_code("c49f6d2c2c74ad34")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

compute_policy_loss Ran
ran; our contract draft was wrong, not the code
StigLidu/GradAlign/select/grpo_loss.py
pointer only (licence: NONE) · get_code("9d79168f0a9a110a")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

describe_target Ran
ran; our contract draft was wrong, not the code
StigLidu/GradAlign/automated/convert_megatron_optimizer_to_hf.py
pointer only (licence: NONE) · get_code("acebf9fd0c7337b5")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

detect_parallel_config Ran
ran, honoured the contract we drafted
StigLidu/GradAlign/automated/convert_megatron_optimizer_to_hf.py
pointer only (licence: NONE) · get_code("d46df850e9d2e176")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

find_part_dirs Ran
ran; our contract draft was wrong, not the code
StigLidu/GradAlign/automated/aggregate.py
pointer only (licence: NONE) · get_code("ce946da43a827e70")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

part_dir_exists Ran
ran, violated the contract we drafted
StigLidu/GradAlign/automated/run_parallel_analysis.py
pointer only (licence: NONE) · get_code("c7cb10df1b7fbd5d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

read_json Ran
ran; our contract draft was wrong, not the code
StigLidu/GradAlign/sort_jsonl.py
pointer only (licence: NONE) · get_code("7c336680bdbe7af1")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

sort_entries Ran
ran; our contract draft was wrong, not the code
StigLidu/GradAlign/sort_jsonl.py
pointer only (licence: NONE) · get_code("bb7ead774bbf8c6f")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

aggregate_accuracy Not yet run
unverified (harvested, no recorded run)
StigLidu/GradAlign/automated/aggregate.py
pointer only (licence: NONE) · get_code("bc620ec0d5ea56a5")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

get_dataset_dir Not yet run
unverified (harvested, no recorded run)
StigLidu/GradAlign/automated/config.py
pointer only (licence: NONE) · get_code("b160cd5de60e562d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

get_model_path Not yet run
unverified (harvested, no recorded run)
StigLidu/GradAlign/automated/config.py
pointer only (licence: NONE) · get_code("99af6d5e08f8669f")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

get_model_type Not yet run
unverified (harvested, no recorded run)
StigLidu/GradAlign/automated/config.py
pointer only (licence: NONE) · get_code("17dc8f1091f17309")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, StigLidu/GradAlign when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 17 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Reinforcement learning (RL) has become a central post-training paradigm for large language models (LLMs), but its performance is highly sensitive to the quality of training problems. This sensitivity stems from the non-stationarity of RL: rollouts are generated by an evolving policy, and learning is shaped by exploration and reward feedback, unlike supervised fine-tuning (SFT) with fixed trajectories. As a result, prior work often relies on manual curation or simple heuristic filters (e.g., accuracy), which can admit incorrect or low-utility problems. We propose GradAlign, a gradient-aligned data selection method for LLM reinforcement learning that uses a small, trusted validation set to prioritize training problems whose policy gradients align with validation gradients, yielding an adaptive curriculum. We evaluate GradAlign across three challenging data regimes: unreliable reward signals, distribution imbalance, and low-utility training corpus, showing that GradAlign consistently outperforms existing baselines, underscoring the importance of directional gradient signals in navigating non-stationary policy optimization and yielding more stable training and improved final performance. We release our implementation at https://github.com/StigLidu/GradAlign.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2602.21492")
get_code_for_paper("2602.21492")
have("2602.21492")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2602.21492.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2602.21492.svg)](https://syntology.ai/paper/2602.21492)

Connect an agent — have() is free.