SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2606.20008 · 2026

VIMPO: Value-Implicit Policy Optimization for LLMs

U Berkeley, Sergey Levine, Xuandong Zhao, Aosong Feng, Zhewei Kang

arXiv · PDF · Open in the Atlas

Code that ran

5 of 7 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; 2 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/2606.20008.json · badge: /api/ran/2606.20008.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
backprop07/VIMPO canonical (classifier label)
classifier basis: census
5 ran · 2 no recorded run
FunctionStatusWhere it lives
build_vimpo_token_term Ran
ran; our contract draft was wrong, not the code
backprop07/VIMPO/recipe/vimpo/vimpo_core_algos.py
code served (permissive licence) · get_code("4f354dd4e5f0d378")
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 Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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)

# Create small tensors with distinct shapes for clarity
batch_size = 2
seq_length = 4

# log_prob: log probabilities from the policy model
# Shape: (batch_size, seq_length)
log_prob = torch.randn(batch_size, seq_length)

# ref_log_prob: log probabilities from the reference model
# Shape: (batch_size, seq_length)
ref_log_prob = torch.randn(batch_size, seq_length)

# token_kl: KL divergence per token
# Shape: (batch_size, seq_length)
token_kl = torch.randn(batch_size, seq_length)

# response_mask: binary mask indicating valid response tokens
# Shape: (batch_size, seq_length)
response_mask = torch.ones(batch_size, seq_length, dtype=torch.bool)

# beta: scaling factor for the VIMPO term
beta = 0.1

ARGS = {
    "log_prob": log_prob,
    "ref_log_prob": ref_log_prob,
    "token_kl": token_kl,
    "response_mask": response_mask,
    "beta": beta,
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
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
compute_grouped_baseline_by_uid Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
backprop07/VIMPO/recipe/vimpo/vimpo_core_algos.py
code served (permissive licence) · get_code("af563e4e391efceb")
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 Tensor, ndim 1, dtype family float, shape (2,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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)

# Create a 1D tensor of values (batch size 2)
values = torch.tensor([1.0, 2.0], dtype=torch.float32)

# Create UIDs - can be tensor scalars or Python objects
# Using two different UIDs so we have two groups
uids = [torch.tensor(0), torch.tensor(1)]

ARGS = {
    'values': values,
    'uids': uids,
}
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 Tensor, ndim 1, dtype family float, shape (2,) — shape (2,) as a later run recorded it
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
generate_config_from_args Ran
ran; our contract draft was wrong, not the code
backprop07/VIMPO/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_dynamic_pipeline_shards Ran
ran, honoured the contract we drafted
backprop07/VIMPO/verl/model_merger/megatron_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("94209884dcb2308e")
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.

positive_int Ran
ran, honoured the contract we drafted
backprop07/VIMPO/recipe/vimpo/data_preprocess/duplicate_aime.py
code served (permissive licence) · get_code("e27925bf449daab9")
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 int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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)

ARGS = {
    'value': '42'
}
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 int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
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
get_weight_loader Not yet run
unverified (harvested, no recorded run)
backprop07/VIMPO/verl/models/weight_loader_registry.py
code served (permissive licence) · get_code("acf240079d434746")
How it ran

no invocation recorded.

get_weight_saver Not yet run
unverified (harvested, no recorded run)
backprop07/VIMPO/verl/models/weight_loader_registry.py
code served (permissive licence) · get_code("ada2d6cf9ca9ee89")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 4 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 4 the same month, 4 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
get_weight_loader 58 other papers
42 dated earlier, 6 the same month, 8 later, 2 undated
2409.19256 (2024-09)
HybridFlow: A Flexible and Efficient RLHF Framework
2505.14362 (2025-05)
DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning
2505.20046 (2025-05)
REARANK: Reasoning Re-ranking Agent via Reinforcement Learning
and 55 more not named here
generate_config_from_args 57 other papers
36 dated earlier, 7 the same month, 11 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
get_dynamic_pipeline_shards 52 other papers
33 dated earlier, 5 the same month, 12 later, 2 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 49 more not named here
get_weight_saver 17 other papers
8 dated earlier, 2 the same month, 7 later
2601.02648 (2026-01)
Prioritized Replay for RL Post-training
2602.04879 (2026-02)
Rethinking the Trust Region in LLM Reinforcement Learning
2602.10609 (2026-02)
Online Causal Kalman Filtering for Stable and Effective Policy Optimization
and 14 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

Reinforcement learning with verifiable rewards has become a central tool for improving the reasoning ability of large language models, but current methods face a trade-off between simplicity and credit assignment. Group-relative methods such as GRPO avoid training a critic, but typically assign a trajectory-level advantage to every token. Actor-critic methods provide denser learning signals, but require a learned value function with its own training instability. We introduce VIMPO, a critic-free policy optimization method that derives a policy-implied value function from the optimality conditions of KL-regularized reinforcement learning. For autoregressive generation, the resulting value recurrence can be written in terms of policy-reference log-ratios and anchored by the terminal condition that no future reward remains at the end of a trajectory. This gives a simple value loss that incorporates outcome-level verifiable rewards without training a critic. The same derivation also yields a critic-free actor advantage, allowing VIMPO to separate reward incorporation through the value loss from policy improvement through a PPO-style actor update. On mathematical RLVR benchmarks, VIMPO improves over GRPO across MATH-500, AIME 2024, AIME 2025, and OlympiadBench, with especially larger gains on competition-style evaluations. Under noisy rewards, VIMPO retains a consistent advantage over GRPO, suggesting that policy-implied value optimization can provide finer credit assignment while preserving the practical simplicity of critic-free training. Code is available at https://github.com/backprop07/VIMPO.

For agents

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

get_harvested_code_for_paper("2606.20008")
get_code_for_paper("2606.20008")
have("2606.20008")

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

curl https://syntology.ai/api/ran/2606.20008.json

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

[![Syntology run record](https://syntology.ai/api/ran/2606.20008.svg)](https://syntology.ai/paper/2606.20008)

Connect an agent — have() is free.