SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2407.08056 · ICLR · 2025

PARETO LOW-RANK ADAPTERS: EFFICIENT MULTI-TASK LEARNING WITH PREFERENCES Nikolaos Dimitriadis EPFL

Pascal Frossard

arXiv · PDF · Open in the Atlas

Code that ran

7 of 11 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (7 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 7 have no contract verdict; 0 failures of our instrument; 4 no recorded run), measured 2026-11-21; 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/2407.08056.json · badge: /api/ran/2407.08056.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
nik-dim/palora community
Papers with Code archive
7 ran · 4 no recorded run
FunctionStatusWhere it lives
BaseCallback Ran
ran on a synthesized input · constructed an object; no result was computed
nik-dim/palora/src/callbacks/methods/palora.py
code served (permissive licence) · get_code("a4b90474b95d1e6b")
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 an instance of its own class, BaseCallback: the run constructed the object and did not call it, so no result was computed. 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)

class BaseCallback:
    """
    Base class for callbacks.
    """

    def __init__(self):
        pass

    def on_before_setup(self, *args, **kwargs):
        pass

    def on_after_setup(self, *args, **kwargs):
        pass

    def on_before_teardown(self, *args, **kwargs):
        pass

    def on_after_teardown(self, *args, **kwargs):
        pass

    def on_before_fit(self, *args, **kwargs):
        pass

    def on_after_fit(self, *args, **kwargs):
        pass

    def on_before_training_epoch(self, *args, **kwargs):
        pass

    def on_after_training_epoch(self, *args, **kwargs):
        pass

    def on_before_eval_epoch(self, *args, **kwargs):
        pass

    def on_after_eval_epoch(self, *args, **kwargs):
        pass

    def on_before_testing_epoch(self, *args, **kwargs):
        pass

    def on_after_testing_epoch(self, *args, **kwargs):
        pass

    def on_before_training_step(self, *args, **kwargs):
        pass

    def on_after_training_step(self, *args, **kwargs):
        pass

    def on_before_backward(self, *args, **kwargs):
        pass

    def on_after_backward(self, *args, **kwargs):
        pass

    def on_before_forward(self, *args, **kwargs):
        pass

    def on_after_forward(self, *args, **kwargs):
        pass

    def on_before_optimizer_step(self, *args, **kwargs):
        pass

    def on_after_optimizer_step(self, *args, **kwargs):
        pass

    def on_before_validation_step(self, *args, **kwargs):
        pass

    def on_after_validation_step(self, *args, **kwargs):
        pass

    def on_before_testing_step(self, *args, **kwargs):
        pass

    def on_after_testing_step(self, *args, **kwargs):
        pass

    def on_before_validating_interpolations(self, *args, **kwargs):
        pass

    def on_after_validating_interpolations(self, *args, **kwargs):
        pass

    def on_before_predicting_interpolations(self, *args, **kwargs):
        pass

    def on_after_predicting_interpolations(self, *args, **kwargs):
        pass

    def on_before_setting_dataloader(self, *args, **kwargs):
        pass

    def on_after_setting_dataloader(self, *args, **kwargs):
        pass


ARGS = {}
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 BaseCallback — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a BaseCallback 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
DumbWrapper Ran
ran on a synthesized input · constructed an object; no result was computed
nik-dim/palora/src/callbacks/methods/palora.py
code served (permissive licence) · get_code("c516f0c3b8bae172")
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 an instance of its own class, DumbWrapper: the run constructed the object and did not call it, so no result was computed. 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)

# Create a minimal inner module
class SimpleModule(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)
    
    def __getitem__(self, key):
        return self.linear

# Create the wrapped module
inner_module = SimpleModule()

ARGS = {
    'module': inner_module
}
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 DumbWrapper — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a DumbWrapper 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
PaConv2d Ran
ran on a synthesized input · constructed an object; no result was computed
nik-dim/palora/src/callbacks/methods/palora.py
code served (permissive licence) · get_code("c8cc9a95bc28e7e3")
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 an instance of its own class, PaConv2d: the run constructed the object and did not call it, so no result was computed. 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
import math

torch.manual_seed(0)

# PaConv2d inherits from PaConvLoRA which inherits from nn.Module and PaLoRALayer
# The __init__ signature is:
# def __init__(self, conv_module, in_channels, out_channels, kernel_size, num_members: int, 
#              r=0, lora_alpha=1, lora_dropout=0.0, merge_weights=True, **kwargs)

# For PaConv2d, it calls super().__init__(nn.Conv2d, *args, **kwargs)
# So the first positional arg is in_channels, second is out_channels, third is kernel_size

in_channels = 3
out_channels = 4
kernel_size = 3
num_members = 2
r = 8
lora_alpha = 16
lora_dropout = 0.1
merge_weights = False
stride = 1
padding = 1
dilation = 1
groups = 1
bias = True

ARGS = {
    'in_channels': in_channels,
    'out_channels': out_channels,
    'kernel_size': kernel_size,
    'num_members': num_members,
    'r': r,
    'lora_alpha': lora_alpha,
    'lora_dropout': lora_dropout,
    'merge_weights': merge_weights,
    'stride': stride,
    'padding': padding,
    'dilation': dilation,
    'groups': groups,
    'bias': bias,
}
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 PaConv2d — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PaConv2d 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
PaConvLoRA Ran
ran on a synthesized input · constructed an object; no result was computed
nik-dim/palora/src/callbacks/methods/palora.py
code served (permissive licence) · get_code("d02616dab157c976")
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 an instance of its own class, PaConvLoRA: the run constructed the object and did not call it, so no result was computed. 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
import math

torch.manual_seed(0)

# The function uses:
# - conv_module: a callable that creates a conv layer (nn.Conv1d, nn.Conv2d, or nn.Conv3d)
# - in_channels: int
# - out_channels: int
# - kernel_size: int
# - num_members: int
# - r: int (rank)
# - lora_alpha: int
# - lora_dropout: float
# - merge_weights: bool
# - **kwargs: additional conv params like stride, padding, dilation, groups, bias

# For forward pass, it needs:
# - x: torch.Tensor (input to conv)
# - ray: torch.Tensor (weights for combining LoRA members)

# Small tensors with distinct dimensions
batch_size = 2
in_channels = 4
out_channels = 3
kernel_size = 5
num_members = 2
r = 2
lora_alpha = 1
lora_dropout = 0.0
merge_weights = True

# Create input tensor for Conv1d
# Shape: (batch, in_channels, length)
x = torch.randn(batch_size, in_channels, 16)

# Ray tensor: shape (num_members,) for scaling each member's contribution
ray = torch.randn(num_members)

# Conv module - using Conv1d as a simple case
conv_module = nn.Conv1d

ARGS = {
    "conv_module": conv_module,
    "in_channels": in_channels,
    "out_channels": out_channels,
    "kernel_size": kernel_size,
    "num_members": num_members,
    "r": r,
    "lora_alpha": lora_alpha,
    "lora_dropout": lora_dropout,
    "merge_weights": merge_weights,
}
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 PaConvLoRA — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PaConvLoRA 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
PaLinear Ran
ran on a synthesized input · constructed an object; no result was computed
nik-dim/palora/src/callbacks/methods/palora.py
code served (permissive licence) · get_code("0b4281cff67777fa")
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 an instance of its own class, PaLinear: the run constructed the object and did not call it, so no result was computed. 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
import math

torch.manual_seed(0)

# Define minimal arguments for PaLinear.__init__
ARGS = {
    'in_features': 4,
    'out_features': 3,
    'num_members': 2,
    'r': 2,
    'lora_alpha': 1,
    'lora_dropout': 0.1,
    'fan_in_fan_out': False,
    'merge_weights': False,
    'bias': True,
}
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 PaLinear — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PaLinear 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
PaLoRALayer Ran
ran on a synthesized input · constructed an object; no result was computed
nik-dim/palora/src/callbacks/methods/palora.py
code served (permissive licence) · get_code("ac4072531d78ed88")
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 an instance of its own class, PaLoRALayer: the run constructed the object and did not call it, so no result was computed. 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)

ARGS = {
    "r": 8,
    "lora_alpha": 16,
    "lora_dropout": 0.1,
    "merge_weights": True,
    "num_members": 4,
}
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 PaLoRALayer — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PaLoRALayer 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
Sampler Ran
ran on a synthesized input · constructed an object; no result was computed
nik-dim/palora/src/callbacks/methods/palora.py
code served (permissive licence) · get_code("8f4dd15fcc2cd0f5")
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 an instance of its own class, Sampler: the run constructed the object and did not call it, so no result was computed. 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 dataclasses

torch.manual_seed(0)

@dataclasses.dataclass
class Sampler:
    num_tasks: int
    num: int

    def sample(self, t: float) -> torch.Tensor:
        raise NotImplementedError

ARGS = {
    "num_tasks": 5,
    "num": 10,
}
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 Sampler — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Sampler 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
AlgoCallback Not yet run
unverified (harvested, no recorded run)
nik-dim/palora/src/callbacks/methods/palora.py
code served (permissive licence) · get_code("c98caa99b682ad0b")
How it ran

no invocation recorded.

PaLoRA Not yet run
unverified (harvested, no recorded run)
nik-dim/palora/src/callbacks/methods/palora.py
code served (permissive licence) · get_code("398523ae51a3f57d")
How it ran

no invocation recorded.

PaSequential Not yet run
unverified (harvested, no recorded run)
nik-dim/palora/src/callbacks/methods/palora.py
code served (permissive licence) · get_code("0c324bf9e69afb36")
How it ran

no invocation recorded.

ParetoFrontApproximationAlgoCallback Not yet run
unverified (harvested, no recorded run)
nik-dim/palora/src/callbacks/methods/palora.py
code served (permissive licence) · get_code("ef65a6c3d3f70a89")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 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 11 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 2026-11-21.

Repositories linked to this paper

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

Abstract

Multi-task trade-offs in machine learning can be addressed via Pareto Front Learning (PFL) methods that parameterize the Pareto Front (PF) with a single model. PFL permits to select the desired operational point during inference, contrary to traditional Multi-Task Learning (MTL) that optimizes for a single trade-off decided prior to training. However, recent PFL methodologies suffer from limited scalability, slow convergence, and excessive memory requirements, while exhibiting inconsistent mappings from preference to objective space. We introduce PaLoRA, a novel parameter-efficient method that addresses these limitations in two ways. First, we augment any neural network architecture with task-specific low-rank adapters and continuously parameterize the PF in their convex hull. Our approach steers the original model and the adapters towards learning general and task-specific features, respectively. Second, we propose a deterministic sampling schedule of preference vectors that reinforces this division of labor, enabling faster convergence and strengthening the validity of the mapping from preference to objective space throughout training. Our experiments show that PaLoRA outperforms state-of-the-art MTL and PFL baselines across various datasets, scales to large networks, reducing the memory overhead 23.8 -31.7 times compared with competing PFL baselines in scene understanding benchmarks.

For agents

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

get_harvested_code_for_paper("2407.08056")
get_code_for_paper("2407.08056")
have("2407.08056")

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

curl https://syntology.ai/api/ran/2407.08056.json

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

[![Syntology run record](https://syntology.ai/api/ran/2407.08056.svg)](https://syntology.ai/paper/2407.08056)

Connect an agent — have() is free.