SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.02546 · 2026

D 2 Quant: Accurate Low-bit Post-Training Weight Quantization for LLMs

Yulun Zhang, Ruobing Xie, Xingwu Sun, Xianglong Yan, Chengzhu Bao, Zhiteng Li, Tianao Zhang, Shaoqiu Zhang, Layernorm Layernorm

arXiv · PDF · Open in the Atlas

Code that ran

11 of 16 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, 1 violated it and 0 have no contract verdict; 9 failures of our instrument; 5 no recorded run); 1 of them was also run on the host, not in the jail, by the peer-agreement lane that set its level, measured 2026-11-22; 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/2602.02546.json · badge: /api/ran/2602.02546.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
XIANGLONGYAN/D2Quant canonical (classifier label)
classifier basis: census
11 ran · 5 no recorded run
FunctionStatusWhere it lives
round_ste Ran
ran; our contract draft was wrong, not the code
XIANGLONGYAN/D2Quant/d2quant/quantization/quantizer.py
code served (permissive licence) · get_code("675da8641c602b03")
How it ran

Ran in Syntology's jail on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. The comparison behind this sample's level is a different run: the peer-agreement lane ran it again on the host, not in the jail (report dated 2026-09-05T00:12:31Z). 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 takes a single parameter: x, which is a torch.Tensor
# It performs rounding with straight-through estimator
# We need a tensor with some values that will be rounded

x = torch.randn(2, 4)

ARGS = {
    "x": x
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) 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
apply_mean_shift Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
XIANGLONGYAN/D2Quant/d2quant/quantization/dac.py
code served (permissive licence) · get_code("c2745ed51515c941")
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 1, dtype family float, shape (4,) 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
import torch.nn as nn

torch.manual_seed(0)

# Create a minimal RMSNorm module
class MinimalRMSNorm(nn.Module):
    def __init__(self, dim: int):
        super().__init__()
        self.weight = nn.Parameter(torch.ones(dim))
    
    def forward(self, x: torch.Tensor) -> torch.Tensor:
        # Simple RMSNorm: x / RMS(x) * weight
        rms = torch.sqrt((x ** 2).mean(dim=-1, keepdim=True) + 1e-6)
        return (x / rms) * self.weight

# Create RMSNorm instance
rms = MinimalRMSNorm(dim=4)

# Create student and teacher output lists
# Each tensor has shape (batch=2, hidden=4)
y_student_list = [
    torch.randn(2, 4, dtype=torch.float32),
    torch.randn(2, 4, dtype=torch.float32),
]

y_teacher_list = [
    torch.randn(2, 4, dtype=torch.float32),
    torch.randn(2, 4, dtype=torch.float32),
]

# Define all arguments
ARGS = {
    'rms': rms,
    'y_student_list': y_student_list,
    'y_teacher_list': y_teacher_list,
    'clip': 10.0,
    'lmbd': 1.0,
}
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 1, dtype family float, shape (4,) — shape (4,) 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
asym_quant Ran
ran; our contract draft was wrong, not the code
XIANGLONGYAN/D2Quant/d2quant/quantization/quantizer.py
code served (permissive licence) · get_code("5daeed014eb71da9")
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 small test tensors
x = torch.randn(2, 4)  # Input tensor to quantize
scale = torch.tensor([0.1])  # Scale factor for quantization
zero = torch.tensor([0], dtype=torch.long)  # Zero point
maxq = 255  # Maximum quantization value (typical for int8)

ARGS = {
    "x": x,
    "scale": scale,
    "zero": zero,
    "maxq": maxq,
}
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
compose_final_perm Ran
ran, violated the contract we drafted
XIANGLONGYAN/D2Quant/d2quant/quantization/gar.py
code served (permissive licence) · get_code("5a7bfb94df03681c")
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 1, dtype family int, shape (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

torch.manual_seed(0)

# Analyze the function:
# - local_perms: indexed by original_group, should be a list/tensor where each element can be added to offset
# - global_perm: a tensor that we call .tolist() on, iterate over, and use as indices into local_perms
# - group_size: scalar integer used to compute offset

# Example setup:
# Let's say we have 2 groups (global_perm has 2 elements)
# Each group has group_size=4 elements
# local_perms should be a list of 2 tensors, each with 4 elements

group_size = 4
num_groups = 2

# global_perm: permutation of group indices [0, 1] -> [1, 0]
global_perm = torch.tensor([1, 0], dtype=torch.long)

# local_perms: list of tensors, one per group
# Each tensor contains a permutation within the group
local_perms = [
    torch.tensor([0, 1, 2, 3], dtype=torch.long),  # local perm for group 0
    torch.tensor([3, 2, 1, 0], dtype=torch.long),  # local perm for group 1
]

ARGS = {
    "local_perms": local_perms,
    "global_perm": global_perm,
    "group_size": group_size,
}
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 1, dtype family int, shape (8,) — shape (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: none by the harness; any seed is in the setup block
compute_global_perm Ran
ran; our contract draft was wrong, not the code
XIANGLONGYAN/D2Quant/d2quant/quantization/gar.py
code served (permissive licence) · get_code("4bcf3cf0ed281cca")
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 1, dtype family int, shape (4,) 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 expects:
# - diag_h: a tensor containing diagonal Hessian values
# - group_size: an integer specifying how many elements per group

# The function:
# 1. Computes num_groups = diag_h.numel() // group_size
# 2. Iterates through groups and computes max per group
# 3. Returns argsort of the metrics in descending order

# Let's create a small tensor with 8 elements and group_size of 2
# This gives us 4 groups
diag_h = torch.tensor([0.1, 0.5, 0.3, 0.2, 0.8, 0.4, 0.6, 0.9], dtype=torch.float32)
group_size = 2

ARGS = {
    "diag_h": diag_h,
    "group_size": group_size,
}
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 1, dtype family int, shape (4,) — shape (4,) 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
compute_local_perms Ran
ran; our contract draft was wrong, not the code
XIANGLONGYAN/D2Quant/d2quant/quantization/gar.py
code served (permissive licence) · get_code("d16524faf29750ef")
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 list, ndim 2, dtype family int; 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

torch.manual_seed(0)

# The function expects:
# - diag_h: a 1D tensor of Hessian diagonal values
# - group_size: an integer specifying how many elements per group

# Create a small diagonal Hessian tensor
# Using 8 elements so we can divide evenly by group_size=2 or 4
diag_h = torch.randn(8)

# Set group_size to 2, so we get 4 groups
group_size = 2

ARGS = {
    "diag_h": diag_h,
    "group_size": group_size,
}
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 list, ndim 2, dtype family int — a list output has no array shape of its own; none is stated
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
fit_dual_scale_2bit Ran
ran; our contract draft was wrong, not the code
XIANGLONGYAN/D2Quant/d2quant/quantization/dsq_2bit.py
code served (permissive licence) · get_code("1b4ea44880a00a67")
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 2D tensor (m=3 rows, n=4 columns)
x = torch.randn(3, 4, dtype=torch.float32)

ARGS = {
    "x": x,
    "iter": 15,
    "eps": 1e-08,
    "do_normalize": False,
    "ridge": 0.0,
}
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
fit_dual_scale_3bit Ran
ran; our contract draft was wrong, not the code
XIANGLONGYAN/D2Quant/d2quant/quantization/dsq_3bit.py
code served (permissive licence) · get_code("2e8023af78a306fa")
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 2D tensor for quantization
# Shape: (m=2, n=4) - 2 rows, 4 columns
x = torch.randn(2, 4, dtype=torch.float32)

ARGS = {
    "x": x,
    "iter": 3,
    "eps": 1e-8,
    "do_normalize": False,
    "ridge": 0.0,
}
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
get_minq_maxq Ran
ran, honoured the contract we drafted
XIANGLONGYAN/D2Quant/d2quant/quantization/quantizer.py
code served (permissive licence) · get_code("56fe9b504cdc9689")
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, ndim 1, dtype family int; 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

torch.manual_seed(0)

ARGS = {
    "bits": 8,
    "sym": 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 tuple, ndim 1, dtype family int — a tuple output has no array shape of its own; none is stated
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
get_rope_function_name Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
XIANGLONGYAN/D2Quant/d2quant/model_utils.py
code served (permissive licence) · get_code("8005fc23ccbe61fb")
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 str. 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 transformers
from transformers.models.llama.modeling_llama import LlamaForCausalLM
from transformers import LlamaConfig

torch.manual_seed(0)

# Create a minimal LlamaForCausalLM model
config = LlamaConfig(
    vocab_size=256,
    hidden_size=64,
    num_hidden_layers=2,
    num_attention_heads=2,
    num_key_value_heads=2,
    intermediate_size=128,
    max_position_embeddings=512,
)

model = LlamaForCausalLM(config)

ARGS = {
    'model': model,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
model_type_extractor Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
XIANGLONGYAN/D2Quant/d2quant/model_utils.py
code served (permissive licence) · get_code("5dcced271ec7bc12")
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 type. 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 transformers
from transformers.models.opt.modeling_opt import OPTForCausalLM
from transformers.models.llama.modeling_llama import LlamaForCausalLM

torch.manual_seed(0)

# Create a minimal LlamaForCausalLM instance
llama_config = transformers.LlamaConfig(
    vocab_size=256,
    hidden_size=32,
    num_hidden_layers=2,
    num_attention_heads=2,
    intermediate_size=64,
)
llama_model = LlamaForCausalLM(llama_config)

ARGS = {
    "model": llama_model
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type type — a type output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a type 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
apply_deviation_aware_correction Not yet run
unverified (harvested, no recorded run)
XIANGLONGYAN/D2Quant/d2quant/quantization/dac.py
code served (permissive licence) · get_code("11ea86283bc189ec")
How it ran

no invocation recorded.

cleanup_memory Not yet run
unverified (harvested, no recorded run)
XIANGLONGYAN/D2Quant/d2quant/config.py
code served (permissive licence) · get_code("58973fc4e2b31774")
How it ran

no invocation recorded.

get_layers Not yet run
unverified (harvested, no recorded run)
XIANGLONGYAN/D2Quant/d2quant/model_utils.py
code served (permissive licence) · get_code("5cfb768c5400311b")
How it ran

no invocation recorded.

log_snr_from_lists Not yet run
unverified (harvested, no recorded run)
XIANGLONGYAN/D2Quant/d2quant/quantization/diagnostics.py
code served (permissive licence) · get_code("54417cda50d297b8")
How it ran

no invocation recorded.

ppl_eval Not yet run
unverified (harvested, no recorded run)
XIANGLONGYAN/D2Quant/d2quant/evaluation.py
code served (permissive licence) · get_code("c63931568d9a0771")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 16 distinct functions with an identity, 4 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 0 the same month, 2 later) and 12 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
round_ste 27 other papers
26 dated earlier, 1 later · 13 only as an assembly brick
2203.05740 (2022-03)
QDrop: Randomly Dropping Quantization for Extremely Low-bit Post-Training Quantization
2204.12322 (2022-04)
RAPQ: Rescuing Accuracy for Power-of-Two Low-bit Post-training Quantization
and 24 more not named here
get_rope_function_name 4 other papers
2 dated earlier, 2 later
2404.00456 (2024-04)
QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs
2511.04063 (2025-11)
DartQuant: Efficient Rotational Distribution Calibration for LLM Quantization
2604.07955 (2026-04)
Rethinking Residual Errors in Compensation-based LLM Quantization
and 1 more not named here
asym_quant 1 other paper
1 dated earlier
2410.09426 (2024-10)
FLATQUANT: Flatness Matters for LLM Quantization
get_minq_maxq 1 other paper
1 dated earlier
2511.04063 (2025-11)
DartQuant: Efficient Rotational Distribution Calibration for LLM Quantization

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

Large language models (LLMs) deliver strong performance, but their high compute and memory costs make deployment difficult in resourceconstrained scenarios. Weight-only post-training quantization (PTQ) is appealing, as it reduces memory usage and enables practical speedup without low-bit operators or specialized hardware. However, accuracy often degrades significantly in weight-only PTQ at sub-4-bit precision, and our analysis identifies two main causes: (1) downprojection matrices are a well-known quantization bottleneck, but maintaining their fidelity often requires extra bit-width; (2) weight quantization induces activation deviations, but effective correction strategies remain underexplored. To address these issues, we propose D 2 Quant, a novel weightonly PTQ framework that improves quantization from both the weight and activation perspectives. On the weight side, we design a Dual-Scale Quantizer (DSQ) tailored to down-projection matrices, with an absorbable scaling factor that significantly improves accuracy without increasing the bit budget. On the activation side, we propose Deviation-Aware Correction (DAC), which incorporates a mean-shift correction within Layer-Norm to mitigate quantization-induced activation distribution shifts. Extensive experiments across multiple LLM families and evaluation metrics show that D 2 Quant delivers superior performance for weight-only PTQ at sub-4-bit precision. The code and models will be available at https:// github.com/XIANGLONGYAN/D2Quant.

For agents

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

get_harvested_code_for_paper("2602.02546")
get_code_for_paper("2602.02546")
have("2602.02546")

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

curl https://syntology.ai/api/ran/2602.02546.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.02546.svg)](https://syntology.ai/paper/2602.02546)

Connect an agent — have() is free.