SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.10693 · 2026

VESPO: Variational Sequence-level Soft Policy Optimization for Off-Policy LLM Training

Lei Huang, Xiang Cheng, Guobin Shen, Xing Yu, Chenxiao Zhao

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 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; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 4 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/2602.10693.json · badge: /api/ran/2602.10693.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
FloyedShen/VESPO canonical (classifier label)
classifier basis: census
4 ran · 0 no recorded run
FunctionStatusWhere it lives
generate_config_from_args Ran
ran; our contract draft was wrong, not the code
FloyedShen/VESPO/verl/model_merger/base_model_merger.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from LeapLabTHU/OPDVR, is pointer only (licence: NONE) · get_code("8896f4e2bafe8abc")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from LeapLabTHU/OPDVR, and GitHub detected no licence in that repository 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.

get_reward_function Ran
ran; our contract draft was wrong, not the code
FloyedShen/VESPO/recipe/vespo/code/reward_function.py
code served (permissive licence) · get_code("f51214308fcd5352")
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
import numpy as np

torch.manual_seed(0)

# Define ARGS for get_reward_function
ARGS = {
    'ability': 'math'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, 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
instruction_following Ran
ran; our contract draft was wrong, not the code
FloyedShen/VESPO/recipe/vespo/code/reward_function.py
code served (permissive licence) · get_code("49cf2d65730fa6b6")
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 dict. 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
import torch.nn as nn

torch.manual_seed(0)

# Construct valid arguments for instruction_following function
solution_str = "The answer is 42"
ground_truth = "The answer is 42"

ARGS = {
    'solution_str': solution_str,
    'ground_truth': ground_truth,
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
math Ran
ran; our contract draft was wrong, not the code
FloyedShen/VESPO/recipe/vespo/code/reward_function.py
code served (permissive licence) · get_code("b1acab41856d80d1")
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 dict. 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
import numpy as np

# Mock the math-verify library functions since they may not be available
def parse(expr, parsing_timeout=30):
    """Mock parse function that returns a simplified representation."""
    if expr is None:
        return None
    # Return a simple string representation
    return str(expr).strip()

def verify(gold, pred, timeout_seconds=30):
    """Mock verify function that compares parsed expressions."""
    if gold is None or pred is None:
        return False
    # Simple string comparison for mock
    return str(gold) == str(pred)

# Inject mocks into the module namespace where math() will look for them
import sys
current_module = sys.modules[__name__]
if not hasattr(current_module, 'parse'):
    current_module.parse = parse
if not hasattr(current_module, 'verify'):
    current_module.verify = verify

torch.manual_seed(0)

ARGS = {
    'data_source': None,
    'solution_str': '${1,2,3,4}$',
    'ground_truth': '${1,3} \\cup {2,4}$',
    'extra_info': None,
    'sandbox_fusion_url': None,
    'concurrent_semaphore': None,
    'memory_limit_mb': None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 1 the same month, 1 later) and 3 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.

FunctionAlso present inEarliest-datedOthers named
generate_config_from_args 57 other papers
8 dated earlier, 9 the same month, 37 later, 3 undated
2410.02992 (2024-10)
Learning to Better Search with Language Models via Guided Reinforced Self-Training
2509.25787 (2025-09)
Self-Evolving Vision-Language Models for Image Quality Assessment via Voting and Ranking
2510.08630 (2025-10)
ExPO-HM: Learning to Explain-then-Detect for Hateful Meme Detection
and 54 more not named here

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

Off-policy updates are inevitable in reinforcement learning (RL) for large language models (LLMs) due to rollout staleness from asynchronous training and mismatches between training and inference engines. Naive importance sampling gives an unbiased correction but suffers from high variance, which is amplified by unbounded ratios and autoregressive generation. Prior remedies either rely on scenario-specific engineering, or trade bias for variance via token-level clipping or sequence-level normalization, yet these approaches remain largely heuristic. We propose Variational sEquence-level Soft Policy Optimization (VESPO). By explicitly incorporating variance reduction into a variational formulation, we derive a principled closed-form reshaping kernel that operates directly on sequence-level importance weights, avoids token-level approximation and length normalization, and admits an explicit variance bound for the deployed kernel. Experiments on math reasoning and code generation show that VESPO maintains stable training under severe off-policy conditions (staleness up to 64×) and delivers consistent gains across both dense and Mixture-of-Experts (MoE) models, outperforming recent reshaping baselines under matched setup. Code is available at https://github.com/FloyedShen/VESPO.

For agents

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

get_harvested_code_for_paper("2602.10693")
get_code_for_paper("2602.10693")
have("2602.10693")

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

curl https://syntology.ai/api/ran/2602.10693.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.10693.svg)](https://syntology.ai/paper/2602.10693)

Connect an agent — have() is free.