SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2108.06098 · ICLR · 2022

FedPara: Low-rank Hadamard Product for Communication-Efficient Federated Learning

Nam Hyeon-Woo, Moon Ye-Bin, Tae-Hyun Oh

arXiv · PDF · Open in the Atlas

Code that ran

10 of 26 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; 1 honoured the contract we drafted, 3 violated it and 1 has no contract verdict; 5 failures of our instrument; 16 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/2108.06098.json · badge: /api/ran/2108.06098.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
kohakublueleaf/lycoris community
Papers with Code archive
10 ran · 16 no recorded run
FunctionStatusWhere it lives
_compile_usable Ran
ran on a synthesized input
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("c1945e2d33dd75a2")
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 bool, ndim 0, 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 os
import torch

# The function _compile_usable takes no parameters
# It checks environment variables and imports, so we just need to ensure
# the environment is set up correctly for the function to run

# Clear any TORCHDYNAMO_DISABLE setting that might interfere
if 'TORCHDYNAMO_DISABLE' in os.environ:
    del os.environ['TORCHDYNAMO_DISABLE']

# Ensure torch is available
torch.manual_seed(0)

# Since _compile_usable() takes no parameters, ARGS should be empty
ARGS = {}
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 bool, ndim 0, dtype family float — no output shape is stated: observe_signature.py could not observe it (no_inputs)
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
_scale Ran
ran, honoured the contract we drafted
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("930aebf19ca4dca8")
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 float, 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)

# The function _scale takes a single parameter 'gamma' which can be either:
# 1. A regular Python float/int that gets converted to float
# 2. A torch.Tensor that gets converted to float via .item()

# Test case 1: gamma as a scalar (float)
gamma_scalar = 3.14

# Test case 2: gamma as a torch.Tensor
gamma_tensor = torch.tensor(2.71)

# We'll use the tensor case as it exercises both branches
ARGS = {
    "gamma": gamma_tensor
}
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 float, 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
cast_operands Ran
ran; our contract draft was wrong, not the code
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("fc584dfaf5a8acf7")
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 list. 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

torch.manual_seed(0)

# Create a list of tensors with different dtypes
# Some will match compute dtype, some won't
tensor_fp32 = torch.randn(2, 4, dtype=torch.float32)
tensor_fp64 = torch.randn(2, 4, dtype=torch.float64)
tensor_int32 = torch.randint(0, 10, (2, 4), dtype=torch.int32)
none_tensor = None

tensors = [tensor_fp32, tensor_fp64, tensor_int32, none_tensor]

# Compute dtype to cast to
compute_dtype = torch.float32

ARGS = {
    "tensors": tensors,
    "compute": compute_dtype,
}
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 list — a list output has no array shape of its own; none is 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
compiled Ran
ran; our contract draft was wrong, not the code
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("686db01715dc4fcc")
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 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 functools

torch.manual_seed(0)

# The function `compiled` takes a single parameter `fn` which is a callable
# that will be compiled with torch.compile. We need to provide a simple function
# that can be compiled.

def simple_fn(x):
    """A simple function that performs basic tensor operations."""
    return x * 2 + 1

ARGS = {
    "fn": simple_fn
}
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 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
half_of Ran
ran; our contract draft was wrong, not the code
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("9aa96c05d80a302c")
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 dtype. 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

torch.manual_seed(0)

# The function takes a single parameter 'dtypes' which should be an iterable of torch.dtype objects
# The function checks if any of these dtypes are in HALF (torch.float16 or torch.bfloat16)
# and raises an error if both appear together

# Test case 1: dtypes with a single 16-bit float type
ARGS = {
    'dtypes': [torch.float32, torch.float16]
}
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 dtype — a dtype output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dtype 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
make_weight_tucker Ran
ran; our contract draft was wrong, not the code
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("6c1c60de009f53a9")
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 3, dtype family float, shape (5, 4, 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)

# Analyze the function signature and einsum operations
# t1, w1d, w1u: first tuple decomposition
# t2, w2d, w2u: second tuple decomposition
# scale: scalar multiplier

# From forward pass:
# rebuild1 = torch.einsum('i j ..., j r, i p -> p r ...', t1, w1d, w1u)
# This means:
# - t1 has shape (i, j, ...)
# - w1d has shape (j, r)
# - w1u has shape (i, p)
# - output has shape (p, r, ...)

# Let's use:
# i=2, j=3, r=4, p=5, and some trailing dimensions (batch-like)
# Use ellipsis dimension size of 2

i, j, r, p = 2, 3, 4, 5
trailing_dims = (2,)

# t1: shape (i, j, *trailing_dims)
t1 = torch.randn(i, j, *trailing_dims)

# w1d: shape (j, r)
w1d = torch.randn(j, r)

# w1u: shape (i, p)
w1u = torch.randn(i, p)

# t2: shape (i, j, *trailing_dims)
t2 = torch.randn(i, j, *trailing_dims)

# w2d: shape (j, r)
w2d = torch.randn(j, r)

# w2u: shape (i, p)
w2u = torch.randn(i, p)

# scale: scalar tensor
scale = torch.tensor(1.0)

ARGS = {
    't1': t1,
    'w1d': w1d,
    'w1u': w1u,
    't2': t2,
    'w2d': w2d,
    'w2u': w2u,
    'scale': scale
}
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 3, dtype family float, shape (5, 4, 2) — shape (5, 4, 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
on_device Ran
ran, violated the contract we drafted
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("5757ad34222c2c4c")
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 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)

# The function checks if all tensors in the input are:
# 1. Actually tensors
# 2. On CUDA devices
# 3. Floating point
# 4. All on the same device

# Since we're CPU-only, we'll create a list of tensors that would satisfy
# the conditions if they were on CUDA. The function will return False
# because they're not on CUDA, but that's the correct behavior for CPU tensors.

# Create a list of floating point tensors (on CPU)
tensor1 = torch.randn(2, 4, dtype=torch.float32)
tensor2 = torch.randn(3, 5, dtype=torch.float32)
tensor3 = torch.randn(2, 3, dtype=torch.float32)

# Also test with None values mixed in
tensors_with_none = [tensor1, None, tensor2]

# Test with non-floating tensors
int_tensor = torch.randint(0, 10, (2, 4))
mixed_types = [tensor1, int_tensor]

# Test with single tensor
single_tensor = [tensor1]

# Test with empty list
empty_list = []

ARGS = {
    "tensors": tensors_with_none
}
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 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
one_family Ran
ran, violated the contract we drafted
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("0bbd3fc4c063623a")
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 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)

# The function checks if tensors contain both fp16 and bf16 dtypes
# It takes a list/iterable of tensors (some may be None)

# Test case 1: Valid case - only fp16 tensors
tensor_fp16_1 = torch.randn(2, 4, dtype=torch.float16)
tensor_fp16_2 = torch.randn(2, 4, dtype=torch.float16)

# Test case 2: Valid case - only bf16 tensors
tensor_bf16_1 = torch.randn(2, 4, dtype=torch.bfloat16)
tensor_bf16_2 = torch.randn(2, 4, dtype=torch.bfloat16)

# Test case 3: Valid case - only fp32 tensors
tensor_fp32 = torch.randn(2, 4, dtype=torch.float32)

# Test case 4: Valid case - mix of fp32 and fp16
tensor_mixed_1 = torch.randn(2, 4, dtype=torch.float32)
tensor_mixed_2 = torch.randn(2, 4, dtype=torch.float16)

# Test case 5: Invalid case - both fp16 and bf16
tensor_invalid_1 = torch.randn(2, 4, dtype=torch.float16)
tensor_invalid_2 = torch.randn(2, 4, dtype=torch.bfloat16)

# Test case 6: Valid case - with None values mixed in
tensor_with_none_1 = torch.randn(2, 4, dtype=torch.float16)
tensor_with_none_2 = None
tensor_with_none_3 = torch.randn(2, 4, dtype=torch.float16)

ARGS = {
    "tensors": [tensor_fp16_1, tensor_fp16_2, None]
}
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 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
restore Ran
ran; our contract draft was wrong, not the code
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("169c00861c4d2f60")
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 list. 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 sample gradients with different dtypes
grad1 = torch.randn(2, 4, dtype=torch.float32)
grad2 = torch.randn(3, 5, dtype=torch.float64)
grad3 = None

grads = [grad1, grad2, grad3]

# Create corresponding target dtypes
dtypes = [torch.float32, torch.float64, None]

ARGS = {
    "grads": grads,
    "dtypes": dtypes,
}
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 list — a list output has no array shape of its own; none is 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
static_scale Ran
ran, violated the contract we drafted
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("3d7a016b112adce3")
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 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)

# Test case 1: gamma is a plain float (not a tensor)
gamma_float = 3.14

# Test case 2: gamma is a tensor without gradient tracking
gamma_tensor_no_grad = torch.tensor([1.0, 2.0, 3.0, 4.0])

# Test case 3: gamma is a tensor with gradient tracking
gamma_tensor_with_grad = torch.tensor([1.0, 2.0, 3.0, 4.0], requires_grad=True)

# The function checks: not (isinstance(gamma, torch.Tensor) and gamma.requires_grad)
# So we need to test the gamma parameter

ARGS = {
    "gamma": gamma_float
}
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 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
HadaRebuildFn Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("c80282e8fc2b17b6")
How it ran

no invocation recorded.

HadaTuckerRebuildFn Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("c1a558c05ff52954")
How it ran

no invocation recorded.

HadaWeight Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("25ce8d9b893fa79d")
How it ran

no invocation recorded.

HadaWeightTucker Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("ff49f45f7cc0dad2")
How it ran

no invocation recorded.

_diff_weight Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("82ead5bb13f48a75")
How it ran

no invocation recorded.

available_backends Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("1e53cb13a60a0c98")
How it ran

no invocation recorded.

call_compiled Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("8d96edb94e53fc58")
How it ran

no invocation recorded.

choose Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("6942530568e5f4cc")
How it ran

no invocation recorded.

demote Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("6ff6c50cfb70d72b")
How it ran

no invocation recorded.

diff_weight Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("4dd9ac3bb6ad0e2b")
How it ran

no invocation recorded.

get_ops Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("c2bf50dc9e8e062e")
How it ran

no invocation recorded.

loha_diff_weight Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("b9fc91af78bb5a55")
How it ran

no invocation recorded.

make_weight Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("42134b9fa83d1b96")
How it ran

no invocation recorded.

promote Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("b2fe3b0f677b2213")
How it ran

no invocation recorded.

resolve Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("af9a28bda7cff15b")
How it ran

no invocation recorded.

resolve_backend Not yet run
unverified (harvested, no recorded run)
kohakublueleaf/lycoris/lycoris/functional/loha.py
code served (permissive licence) · get_code("6b5371eae9399500")
How it ran

no invocation recorded.

Also present in other papers' code

Of 26 samples lifted from this paper's linked repositories, 26 carry a function identity and 0 do not. Of the 26 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 26 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 this work, we propose a communication-efficient parameterization, FedPara, for federated learning (FL) to overcome the burdens on frequent model uploads and downloads. Our method re-parameterizes weight parameters of layers using low-rank weights followed by the Hadamard product. Compared to the conventional low-rank parameterization, our FedPara method is not restricted to lowrank constraints, and thereby it has a far larger capacity. This property enables to achieve comparable performance while requiring 3 to 10 times lower communication costs than the model with the original layers, which is not achievable by the traditional low-rank methods. The efficiency of our method can be further improved by combining with other efficient FL optimizers. In addition, we extend our method to a personalized FL application, pFedPara, which separates parameters into global and local ones. We show that pFedPara outperforms competing personalized FL methods with more than three times fewer parameters. Project

For agents

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

get_harvested_code_for_paper("2108.06098")
get_code_for_paper("2108.06098")
have("2108.06098")

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

curl https://syntology.ai/api/ran/2108.06098.json

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

[![Syntology run record](https://syntology.ai/api/ran/2108.06098.svg)](https://syntology.ai/paper/2108.06098)

Connect an agent — have() is free.