SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2510.04019 · ICML · 2025

Simple Policy Gradients for Reasoning with Diffusion Language Models

Anthony Zhan

arXiv · PDF · Open in the Atlas

Code that ran

5 of 5 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, 0 violated it and 0 have no contract verdict; 3 failures of our instrument; 0 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/2510.04019.json · badge: /api/ran/2510.04019.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
probablyabot/agrpo canonical (classifier label)
classifier basis: census
5 ran · 0 no recorded run
FunctionStatusWhere it lives
answer_format_reward Ran
ran, honoured the contract we drafted
probablyabot/agrpo/rewards.py
code served (permissive licence) · get_code("8c5e4dcd27a0e0d4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re
from typing import List

ANSWER_TAG = re.compile('<answer>\\s*(.*?)\\s*</answer>', re.DOTALL | re.IGNORECASE)

def answer_format_reward(completions: List[str], **kwargs):
    rewards = []
    for comp in completions:
        rewards.append(0.5 * bool(ANSWER_TAG.search(comp)))
    return rewards


# Construct test arguments
completions = [
    "This is a response with <answer>correct answer</answer>",
    "This is a response without the proper tags",
    "<ANSWER>another valid answer</ANSWER>",
    "No answer here at all"
]

ARGS = {
    "completions": completions
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
extract_boxed Ran
ran; our contract draft was wrong, not the code
probablyabot/agrpo/rewards.py
code served (permissive licence) · get_code("4eb074484f3ce4a2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re

def extract_boxed(text: str):
    if (match := re.search('\\\\boxed\\{', text)):
        i = match.end()
        depth = 1
        for j in range(i, len(text)):
            if text[j] == '{':
                depth += 1
            elif text[j] == '}':
                depth -= 1
                if depth == 0:
                    return text[i:j]
    return None

# Test cases for the extract_boxed function
ARGS = {
    "text": "The answer is \\boxed{42}"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
gumbel_sample Ran
ran; our contract draft was wrong, not the code
probablyabot/agrpo/generate.py
code served (permissive licence) · get_code("a7c4e85e89c9c067")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

class SamplingConfig:
    def __init__(self, temperature=1.0, top_k=None, top_p=1.0):
        self.temperature = temperature
        self.top_k = top_k
        self.top_p = top_p

# Create small logits tensor: (batch=2, sequence_length=3, vocab=4)
logits = torch.randn(2, 3, 4)

# Create config with various sampling parameters
config = SamplingConfig(temperature=0.8, top_k=2, top_p=0.9)

ARGS = {
    "logits": logits,
    "config": config,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
register_dataset Ran
ran; our contract draft was wrong, not the code
probablyabot/agrpo/data.py
code served (permissive licence) · get_code("e30ee20f00809d83")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function register_dataset takes one parameter: alias
# It's a string that serves as a key in the DATASETS dictionary
# The function returns a decorator that takes a function and registers it

alias_value = "my_dataset"

ARGS = {
    "alias": alias_value
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
thinking_format_reward Ran
ran, honoured the contract we drafted
probablyabot/agrpo/rewards.py
code served (permissive licence) · get_code("a5ef0b84542283a7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re
from typing import List

THINKING_TAG = re.compile('<think>\\s*(.*?)\\s*</think>', re.DOTALL | re.IGNORECASE)

def thinking_format_reward(completions: List[str], **kwargs):
    rewards = []
    for comp in completions:
        rewards.append(0.5 * bool(THINKING_TAG.search(comp)))
    return rewards


# Setup code
completions = [
    "This is a regular completion without thinking tags.",
    "<think>This is a thinking block with some reasoning</think> And then the answer.",
    "<THINK>Case insensitive thinking tag</THINK> with answer",
    "No thinking here at all",
    "<think>  \n  Whitespace handling in thinking  \n  </think> Final response",
]

ARGS = {
    "completions": completions,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 5 samples lifted from this paper's linked repositories, 5 carry a function identity and 0 do not. Of the 5 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 5 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

Diffusion large language models (dLLMs) represent a promising alternative to autoregressive LLMs; however, the lack of effective post-training techniques, including reinforcement learning (RL), remains a key challenge for dLLMs, especially for downstream applications. Existing approaches often rely on a sequence-level view that requires biased likelihood approximations. In this work, we propose Amortized Group Relative Policy Optimization (AGRPO), a policy gradient algorithm that leverages the Markovian nature of dLLMs, optimizing individual denoising steps rather than full sequences. Our approach improves alignment between the trained policy and the inference process and also admits efficient, unbiased gradient updates via a novel timestep estimation scheme. We demonstrate AGRPO's effectiveness on different math and reasoning tasks, achieving absolute accuracy gains of +59.4% and +69.7% on Countdown and Sudoku over the base LLaDA model, exceeding comparable methods such as diffu-GRPO. Code is available at https://github.com/probablyabot/agrpo.

For agents

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

get_harvested_code_for_paper("2510.04019")
get_code_for_paper("2510.04019")
have("2510.04019")

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

curl https://syntology.ai/api/ran/2510.04019.json

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

[![Syntology run record](https://syntology.ai/api/ran/2510.04019.svg)](https://syntology.ai/paper/2510.04019)

Connect an agent — have() is free.