SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.05367 · ICML · 2026

RaBiT: Residual-Aware Binarization Training for Accurate and Efficient LLMs

Banseok Lee, Dongkyu Kim, Youngcheon You, Youngmin Kim, Minseop Choi, Seonyoung Kim, Hyochan Chong, Changdong Kim

arXiv · PDF · Open in the Atlas

Code that ran

5 of 6 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (1 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 1 has no contract verdict; 4 failures of our instrument; 1 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/2602.05367.json · badge: /api/ran/2602.05367.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
SamsungLabs/RaBiT community
Papers with Code archive
5 ran · 1 no recorded run
FunctionStatusWhere it lives
RandomizedSVD Ran
ran; our contract draft was wrong, not the code
SamsungLabs/RaBiT/quantization/modules/rabit.py
code served (permissive licence) · get_code("68cd777a2535b9fc")
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 tuple. 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)

# Construct a small matrix A for the RandomizedSVD function
# The function expects A to be a 2D tensor that supports @ (matrix multiplication)
# and .shape, .device, .dtype attributes
A = torch.randn(5, 4)  # 5x4 matrix (m=5, n=4)

# k is the number of singular vectors/values to keep
k = 2

# p is the oversampling parameter
p = 10

# q is the number of power iterations
q = 2

ARGS = {
    "A": A,
    "k": k,
    "p": p,
    "q": q
}
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 tuple — a tuple 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
_STENResidualBinarization Ran
ran on a synthesized input · constructed an object; no result was computed
SamsungLabs/RaBiT/quantization/modules/rabit.py
code served (permissive licence) · get_code("a645ce3201ab1568")
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, _STENResidualBinarization: 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)

# Define tensors with distinct dimensions for clarity
# x: input tensor (batch_size=2, input_dim=4)
x = torch.randn(2, 4)

# weight: weight tensor (output_dim=3, input_dim=4)
weight = torch.randn(3, 4)

# Create scale tensors for N-path residual binarization
# Using 2 paths as an example: inscale1, outscale1, inscale2, outscale2
inscale1 = torch.ones(4) * 0.5  # inscale for path 1
outscale1 = torch.ones(3) * 0.5  # outscale for path 1
inscale2 = torch.ones(4) * 0.3  # inscale for path 2
outscale2 = torch.ones(3) * 0.3  # outscale for path 2

ARGS = {
    'x': x,
    'weight': weight,
    'scales': (inscale1, outscale1, inscale2, outscale2)
}
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 _STENResidualBinarization — 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 _STENResidualBinarization 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
binary_packer Ran
ran; our contract draft was wrong, not the code
SamsungLabs/RaBiT/quantization/modules/rabit.py
code served (permissive licence) · get_code("f28711d7e5b0e4ba")
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 int, shape (2, 1) 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 small int8 tensor with values -1 and +1
# Shape: (2, 5) - 2 rows, 5 columns
tensor = torch.tensor([
    [1, -1, 1, -1, 1],
    [-1, 1, -1, 1, -1]
], dtype=torch.int8)

ARGS = {
    "tensor": tensor
}
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 int, shape (2, 1) — shape (2, 1) 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
cyclic_n_svid Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
SamsungLabs/RaBiT/quantization/modules/rabit.py
code served (permissive licence) · get_code("900ce11c43a09b28")
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 tuple. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 small test tensor
# cyclic_n_svid takes w (tensor), dtype, iters, and n as parameters
w = torch.randn(4, 3)  # Small 4x3 tensor
dtype = torch.float32

ARGS = {
    "w": w,
    "dtype": dtype,
    "iters": 2,
    "n": 3
}
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 tuple — a tuple 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
svid_random Ran
ran; our contract draft was wrong, not the code
SamsungLabs/RaBiT/quantization/modules/rabit.py
code served (permissive licence) · get_code("f0670693a31c0ae6")
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 tuple. 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 small weight tensor (matrix) for testing
# Shape: (4, 3) - distinct dimensions for clarity
w = torch.randn(4, 3, dtype=torch.float32)

# dtype parameter
dtype = torch.float32

ARGS = {
    "w": w,
    "dtype": dtype,
}
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 tuple — a tuple 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
RaBiTLinear Not yet run
unverified (harvested, no recorded run)
SamsungLabs/RaBiT/quantization/modules/rabit.py
code served (permissive licence) · get_code("d5a0607082f36021")
How it ran

no invocation recorded.

Also present in other papers' code

Of 6 samples lifted from this paper's linked repositories, 6 carry a function identity and 0 do not. Of the 6 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 5 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
binary_packer
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2506.13771 (2025-06)
LittleBit: Ultra Low-Bit Quantization via Latent Factorization

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

Efficient deployment of large language models (LLMs) requires extreme quantization, forcing a critical trade-off between low-bit efficiency and performance. Residual binarization enables hardware-friendly, matmul-free inference by stacking binary (±1) layers, but is plagued by pathological feature co-adaptation. We identify a key failure mode, which we term inter-path adaptation: during quantization-aware training (QAT), parallel residual binary paths learn redundant features, degrading the error-compensation structure and limiting the expressive capacity of the model. While prior work relies on heuristic workarounds (e.g., path freezing) that constrain the solution space, we propose RaBiT, a novel quantization framework that resolves co-adaptation by algorithmically enforcing a residual hierarchy. Its core mechanism sequentially derives each binary path from a single shared full-precision weight, which ensures that every path corrects the error of the preceding one. This process is stabilized by a robust initialization that prioritizes functional preservation over mere weight approximation. RaBiT redefines the 2-bit accuracyefficiency frontier: it achieves state-of-the-art performance, rivals even hardware-intensive Vector Quantization (VQ) methods, and delivers a 4.49× inference speed-up over full-precision models on an RTX 4090. Code is available at github.com/SamsungLabs/RaBiT

For agents

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

get_harvested_code_for_paper("2602.05367")
get_code_for_paper("2602.05367")
have("2602.05367")

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

curl https://syntology.ai/api/ran/2602.05367.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.05367.svg)](https://syntology.ai/paper/2602.05367)

Connect an agent — have() is free.