SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2501.19400 · ICML · 2025

Vintix: Action Model via In-Context Reinforcement Learning

Vladislav Kurenkov, J Wang, Alexander Nikulin, Denis Tarasov, M Chen, Ilya Zisman, M Sharma, M Kim, K Lin, M Castro, Andrei Polubarov, Nikita Lyubaykin, and 1 more

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; 0 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 2 failures of our instrument; 2 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; 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/2501.19400.json · badge: /api/ran/2501.19400.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
dunnolab/vintix canonical (classifier label)
classifier basis: classifier_2025
5 ran · 2 no recorded run
FunctionStatusWhere it lives
ContinuousAcsEncoder Ran
ran on a synthesized input
dunnolab/vintix/vintix/nn/individual_task_encoder.py
code served (permissive licence) · get_code("81e363a4d42a6acf")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, dtype family float, shape (3, 8) 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
from torch import nn

INIT_ARGS = {
    'group_metadata': {
        'action_dim': 4,
        'action_type': 'continuous'
    },
    'emb_dim': 8,
    'emb_activation': nn.LeakyReLU()
}

ARGS = {
    'acs': torch.randn(3, 4),
    'shift_padding': torch.randn(1, 8)
}

if __name__ == '__main__':
    from typing import Any, Dict, List, Optional
    
    class ContinuousAcsEncoder(nn.Module):
        """Continuous action encoder

        Encode continuous actions with MLP

        Args:
            group_metadata: group-specific metadata
            emb_dim: output embedding dimension
            emb_activation: output activation
        """

        def __init__(self, group_metadata: Dict[str, Any], emb_dim: int, emb_activation: nn.Module=nn.LeakyReLU()):
            super(ContinuousAcsEncoder, self).__init__()
            self.emb_dim = emb_dim
            self.acs_dim = group_metadata['action_dim']
            assert group_metadata['action_type'] == 'continuous'
            self.layers = nn.Sequential(nn.Linear(self.acs_dim, self.emb_dim), emb_activation, nn.LayerNorm(self.emb_dim), nn.Linear(self.emb_dim, self.emb_dim), emb_activation, nn.LayerNorm(self.emb_dim))

        def forward(self, acs: torch.Tensor, shift_padding: Optional[torch.Tensor]=None) -> torch.Tensor:
            """Encode actions

            Action encoding forward pass

            Args:
                acs: continuous actions trajectory with shape
                    (traj_len, acs_dim) and dtype float
                shift_padding: if provided uses its value for
                1 step forward padding for correct SAR triplets
                alignment
            Returns:
                torch.Tensor: encoded continuous actions
            """
            acs_emb = self.layers(acs)
            if shift_padding is not None:
                acs_emb = torch.cat([shift_padding, acs_emb[:-1]], dim=0)
            return acs_emb

    encoder = ContinuousAcsEncoder(**INIT_ARGS)
    output = encoder(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output dtype: {output.dtype}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 (3, 8) — shape (3, 8) 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
LinearObsEncoder Ran
ran on a synthesized input
dunnolab/vintix/vintix/nn/individual_task_encoder.py
code served (permissive licence) · get_code("94fe9914b4618cec")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, dtype family float, shape (2, 8) 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
from torch import nn

# Define the class
class LinearObsEncoder(nn.Module):
    """Continuous observation encoder

    Encoder for proprioceptive observations

    Args:
        group_metadata: task-specific metadata
        emb_dim: output embedding dimension
        emb_activation: output activation
    """

    def __init__(self, group_metadata: dict, emb_dim: int, emb_activation: nn.Module=None):
        super(LinearObsEncoder, self).__init__()
        if emb_activation is None:
            emb_activation = nn.LeakyReLU()
        self.emb_dim = emb_dim
        self.obs_shape = group_metadata['observation_shape']['proprio']
        assert len(self.obs_shape) == 1
        self.layers = nn.Sequential(
            nn.Linear(self.obs_shape[0], self.emb_dim),
            emb_activation,
            nn.LayerNorm(self.emb_dim),
            nn.Linear(self.emb_dim, self.emb_dim),
            emb_activation,
            nn.LayerNorm(self.emb_dim)
        )

    def forward(self, obs: torch.Tensor) -> torch.Tensor:
        """Obs encoder forward pass"""
        return self.layers(obs)


# Setup arguments for __init__
INIT_ARGS = {
    'group_metadata': {
        'observation_shape': {
            'proprio': (4,)
        }
    },
    'emb_dim': 8,
    'emb_activation': nn.LeakyReLU()
}

# Setup arguments for forward pass
ARGS = {
    'obs': torch.randn(2, 4)
}

# Construct and call
if __name__ == '__main__':
    encoder = LinearObsEncoder(**INIT_ARGS)
    output = encoder(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Expected shape: (2, 8)")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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, 8) — shape (2, 8) 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
RewardEncoder Ran
ran on a synthesized input
dunnolab/vintix/vintix/nn/individual_task_encoder.py
code served (permissive licence) · get_code("cd7f80d9a35a3e3a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, 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 torch
import torch.nn as nn

# INIT_ARGS for RewardEncoder.__init__
INIT_ARGS = {
    'emb_dim': 8,
    'emb_activation': nn.LeakyReLU()
}

# ARGS for RewardEncoder.forward
# rews should have shape (traj_len, 1) but will be unsqueezed to (traj_len, 1, 1) in forward
# So we pass shape (traj_len,) which gets unsqueezed to (traj_len, 1) in the method
traj_len = 5

ARGS = {
    'rews': torch.randn(traj_len),  # scalar rewards trajectory
    'shift_padding': torch.randn(1, 8)  # padding with shape (1, emb_dim)
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
get_acs_encoder Ran
ran; our contract draft was wrong, not the code
dunnolab/vintix/vintix/nn/individual_task_encoder.py
code served (permissive licence) · get_code("21d4e50f8ea1c82c")
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) and returned ContinuousAcsEncoder. 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
from torch import nn

torch.manual_seed(0)

# Test with continuous action type
group_metadata_continuous = {
    'action_type': 'continuous',
    'action_dim': 4
}

emb_dim = 8
emb_activation = nn.LeakyReLU()

ARGS = {
    'group_metadata': group_metadata_continuous,
    'emb_dim': emb_dim,
    'emb_activation': emb_activation
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ContinuousAcsEncoder — a ContinuousAcsEncoder output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ContinuousAcsEncoder 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
get_obs_encoder Ran
ran; our contract draft was wrong, not the code
dunnolab/vintix/vintix/nn/individual_task_encoder.py
code served (permissive licence) · get_code("9e20e63115567301")
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 LinearObsEncoder. 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
from torch import nn

torch.manual_seed(0)

# Create group_metadata with proprioceptive observation shape
group_metadata = {
    'observation_shape': {
        'proprio': (7,)  # 1D tuple as expected by LinearObsEncoder
    }
}

# Define embedding dimension
emb_dim = 16

# Define embedding activation
emb_activation = nn.LeakyReLU()

# Define a minimal image encoder (not used in this case, but required as parameter)
class MinimalImageEncoder(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(3, 16)
    
    def forward(self, x):
        return self.linear(x)

image_encoder = MinimalImageEncoder()

ARGS = {
    'group_metadata': group_metadata,
    'emb_dim': emb_dim,
    'emb_activation': emb_activation,
    'image_encoder': image_encoder
}
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 LinearObsEncoder — a LinearObsEncoder output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a LinearObsEncoder 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
DiscreteAcsEncoder Not yet run
unverified (harvested, no recorded run)
dunnolab/vintix/vintix/nn/individual_task_encoder.py
code served (permissive licence) · get_code("4d5af6687b20aa67")
How it ran

no invocation recorded.

IndividualTaskEncoderNew Not yet run
unverified (harvested, no recorded run)
dunnolab/vintix/vintix/nn/individual_task_encoder.py
code served (permissive licence) · get_code("b8a3da843aa83b77")
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, 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 7 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

In-Context Reinforcement Learning (ICRL) represents a promising paradigm for developing generalist agents that learn at inference time through trial-and-error interactions, analogous to how large language models adapt contextually, but with a focus on reward maximization. However, the scalability of ICRL beyond toy tasks and single-domain settings remains an open challenge. In this work, we present the first steps toward scaling ICRL by introducing a fixed, cross-domain model capable of learning behaviors through in-context reinforcement learning. Our results demonstrate that Algorithm Distillation, a framework designed to facilitate ICRL, offers a compelling and competitive alternative to expert distillation to construct versatile action models. These findings highlight the potential of ICRL as a scalable approach for generalist decision-making systems. Code to be released at dunnolab/vintix.

For agents

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

get_harvested_code_for_paper("2501.19400")
get_code_for_paper("2501.19400")
have("2501.19400")

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

curl https://syntology.ai/api/ran/2501.19400.json

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

[![Syntology run record](https://syntology.ai/api/ran/2501.19400.svg)](https://syntology.ai/paper/2501.19400)

Connect an agent — have() is free.