SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2410.07170 · NeurIPS · 2025

Parameter Efficient Fine-tuning via Explained Variance Adaptation

Marc Deisenroth, Sepp Hochreiter, Thomas Schmied, Fabian Paischer, Benedikt Alkin, Lukas Hauzenberger

arXiv · PDF · Open in the Atlas

Code that ran

3 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, 1 violated it and 1 has no contract verdict; 1 failure of our instrument; 1 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/2410.07170.json · badge: /api/ran/2410.07170.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
ml-jku/EVA marked official in Papers with Code
Papers with Code archive
2 ran · 0 no recorded run
BenediktAlkin/vtab1k-pytorch marked official in Papers with Code
Papers with Code archive
1 ran · 1 no recorded run
FunctionStatusWhere it lives
_BufferedIncrementalPCA Ran
ran on a synthesized input
BenediktAlkin/vtab1k-pytorch/src/vtab/adapters/eva_linear.py
code served (permissive licence) · get_code("1da02603acd65cf8")
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 NoneType. 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 sklearn.decomposition import IncrementalPCA, PCA

class _BufferedIncrementalPCA(IncrementalPCA):
    """
    IncrementalPCA that accumulates batches if number of
    samples for partial_fit is smaller than num_components
    """

    def __init__(self, n_components, batch_size):
        super().__init__(n_components=n_components, batch_size=batch_size)
        self.buffer = []
        self.buffer_size = 0

    def partial_fit(self, x):
        assert torch.is_tensor(x) and x.ndim == 2
        x = x.cpu()
        self.buffer.append(x)
        self.buffer_size += len(x)
        if self.buffer_size >= self.batch_size and self.buffer_size >= self.n_components:
            x = torch.concat(self.buffer).numpy()
            assert len(x) > 0
            super().partial_fit(x)
            self.buffer.clear()
            self.buffer_size = 0


INIT_ARGS = {
    'n_components': 3,
    'batch_size': 2,
}

ARGS = {
    'x': torch.randn(2, 4),
}

METHOD = 'partial_fit'


if __name__ == '__main__':
    pca = _BufferedIncrementalPCA(**INIT_ARGS)
    pca.partial_fit(ARGS['x'])
    print("Success: partial_fit executed")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
sklearn, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type NoneType — a NoneType output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a NoneType 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
get_eva_state_dict Ran
ran; our contract draft was wrong, not the code
ml-jku/EVA/src/utils.py
code served (permissive licence) · get_code("66d79fedd060dc95")
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 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
from collections import defaultdict, Counter

torch.manual_seed(0)

# Create a minimal svd_dict with tensors
# The function accesses .size(0) on the first value, so we need tensors
# Each value should be a tensor where we can slice it with [:v]
svd_dict = {
    'layer.0.weight': torch.randn(8, 4),  # 8 components, 4 features
    'layer.1.weight': torch.randn(8, 4),  # 8 components, 4 features
    'layer.2.weight': torch.randn(8, 4),  # 8 components, 4 features
}

# target_modules to match against the keys
target_modules = ['layer.0.weight', 'layer.1.weight', 'layer.2.weight']

# rank parameter
rank = 2

# redistribute parameter
redistribute = False

# exp_vars - optional dict with structure {key: {metric: list_of_values}}
# Only used when redistribute=True, but we'll provide it anyway
exp_vars = {
    'layer.0.weight': {
        'ratio': [0.9, 0.8, 0.7, 0.6, 0.5, 0.4, 0.3, 0.2],
        'raw': [100, 90, 80, 70, 60, 50, 40, 30],
        'sum': [500, 450, 400, 350, 300, 250, 200, 150],
        'max': [100, 95, 90, 85, 80, 75, 70, 65],
    },
    'layer.1.weight': {
        'ratio': [0.85, 0.75, 0.65, 0.55, 0.45, 0.35, 0.25, 0.15],
        'raw': [95, 85, 75, 65, 55, 45, 35, 25],
        'sum': [475, 425, 375, 325, 275, 225, 175, 125],
        'max': [95, 90, 85, 80, 75, 70, 65, 60],
    },
    'layer.2.weight': {
        'ratio': [0.88, 0.78, 0.68, 0.58, 0.48, 0.38, 0.28, 0.18],
        'raw': [98, 88, 78, 68, 58, 48, 38, 28],
        'sum': [490, 440, 390, 340, 290, 240, 190, 140],
        'max': [98, 93, 88, 83, 78, 73, 68, 63],
    },
}

# redist_metric parameter
redist_metric = 'ratio'

# rho parameter
rho = 2

ARGS = {
    'svd_dict': svd_dict,
    'target_modules': target_modules,
    'rank': rank,
    'redistribute': redistribute,
    'exp_vars': exp_vars,
    'redist_metric': redist_metric,
    'rho': rho,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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
match_module_name Ran
ran, violated the contract we drafted
ml-jku/EVA/src/utils.py
code served (permissive licence) · get_code("d82feb60c9bc01b7")
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 bool, ndim 0, dtype family float, 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

torch.manual_seed(0)

# Analyzing the function:
# - module_name: a string that will be split by '.' and sliced
# - name_to_match: a string used to count '.' characters and compared for equality

ARGS = {
    "module_name": "layer1.conv.weight",
    "name_to_match": "conv.weight"
}
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 bool, ndim 0, dtype family float, 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
EvaLinear Not yet run
unverified (harvested, no recorded run)
BenediktAlkin/vtab1k-pytorch/src/vtab/adapters/eva_linear.py
code served (permissive licence) · get_code("9ca1d5db1b9dc592")
How it ran

no invocation recorded.

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, 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 4 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

Foundation models (FMs) are pre-trained on large-scale datasets and then finetuned for a specific downstream task. The most common fine-tuning method is to update pretrained weights via low-rank adaptation (LoRA). Existing initialization strategies for LoRA often rely on singular value decompositions (SVD) of gradients or weight matrices. However, they do not provably maximize the expected gradient signal, which is critical for fast adaptation. To this end, we introduce Explained Variance Adaptation (EVA), an initialization scheme that uses the directions capturing the most activation variance, provably maximizing the expected gradient signal and accelerating fine-tuning. EVA performs incremental SVD on minibatches of activation vectors and selects the right-singular vectors for initialization once they converged. Further, by selecting the directions that capture the most activation-variance for a given rank budget, EVA accommodates adaptive ranks that reduce the number of trainable parameters. We apply EVA to a variety of fine-tuning tasks as language generation and understanding, image classification, and reinforcement learning. EVA exhibits faster convergence than competitors and achieves the highest average score across a multitude of tasks per domain while reducing the number of trainable parameters through rank redistribution. In summary, EVA establishes a new Pareto frontier compared to existing LoRA initialization schemes in both accuracy and efficiency. * Equal contribution 39th Conference on Neural Information Processing Systems (NeurIPS 2025).

For agents

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

get_harvested_code_for_paper("2410.07170")
get_code_for_paper("2410.07170")
have("2410.07170")

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

curl https://syntology.ai/api/ran/2410.07170.json

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

[![Syntology run record](https://syntology.ai/api/ran/2410.07170.svg)](https://syntology.ai/paper/2410.07170)

Connect an agent — have() is free.