SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2605.04084 · 2026

FASQ: Flexible Accelerated Subspace Quantization for Calibration-Free LLM Compression

Yian Wang, Hyoukjun Kwon, Sitao Huang, Ye Qiao, Zhiheng Chen

arXiv · PDF · Open in the Atlas

Code that ran

5 of 7 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, 0 violated it and 0 have no contract verdict; 5 failures of our instrument; 2 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/2605.04084.json · badge: /api/ran/2605.04084.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
mit-han-lab/smoothquant community
Papers with Code archive
4 ran · 1 no recorded run
ist-daslab/gptq community
Papers with Code archive
1 ran · 1 no recorded run
FunctionStatusWhere it lives
quantize Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ist-daslab/gptq/quant.py
code served (permissive licence) · get_code("41c57e55d21c4d96")
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 float, shape (2, 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)

# Create a small tensor for x (batch_size=2, features=4)
x = torch.randn(2, 4)

# Create scalar scale parameter
scale = torch.tensor(0.1)

# Create scalar zero parameter
zero = torch.tensor(0.0)

# Create integer maxq parameter (positive value for normal quantization)
maxq = 255

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 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
quantize_activation_per_tensor_absmax Ran
ran; our contract draft was wrong, not the code
mit-han-lab/smoothquant/smoothquant/fake_quant.py
code served (permissive licence) · get_code("aa2bd5edebc9b6ff")
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 float, shape (2, 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)

# Create a small tensor for quantization
# Shape: (2, 4) - batch size 2, feature dimension 4
t = torch.randn(2, 4)

ARGS = {
    "t": t,
    "n_bits": 8
}
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 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
quantize_activation_per_token_absmax Ran
ran; our contract draft was wrong, not the code
mit-han-lab/smoothquant/smoothquant/fake_quant.py
code served (permissive licence) · get_code("c1cb961b4f5cb75c")
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 (2, 3, 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)

# Create a small activation tensor
# Shape: (batch=2, sequence_length=3, hidden_dim=4)
t = torch.randn(2, 3, 4)

ARGS = {
    "t": t,
    "n_bits": 8,
}
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 (2, 3, 4) — shape (2, 3, 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
quantize_weight_per_channel_absmax Ran
ran; our contract draft was wrong, not the code
mit-han-lab/smoothquant/smoothquant/fake_quant.py
code served (permissive licence) · get_code("770da8497b2ff4ba")
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 float, shape (2, 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 quantize_weight_per_channel_absmax takes:
# - w: a tensor that will have operations like .abs(), .max(), .div_(), .round_(), .mul_() called on it
# - n_bits: an integer for bit width (default 8)

# Create a small weight tensor with shape (2, 4) - 2 channels, 4 elements per channel
w = torch.randn(2, 4)

ARGS = {
    "w": w,
    "n_bits": 8
}
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 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
quantize_weight_per_tensor_absmax Ran
ran; our contract draft was wrong, not the code
mit-han-lab/smoothquant/smoothquant/fake_quant.py
code served (permissive licence) · get_code("dd3493503cb95888")
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 float, shape (2, 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)

# Create a small weight tensor for quantization
# The function uses: w.abs().max(), w.div_(), w.round_(), w.mul_()
# So w needs to be a floating point tensor
w = torch.randn(2, 4, dtype=torch.float32)

ARGS = {
    "w": w,
    "n_bits": 8
}
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 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
Quantizer Not yet run
unverified (harvested, no recorded run)
ist-daslab/gptq/quant.py
code served (permissive licence) · get_code("fc31370a0c6e1aaf")
How it ran

no invocation recorded.

W8A8Linear Not yet run
unverified (harvested, no recorded run)
mit-han-lab/smoothquant/smoothquant/fake_quant.py
code served (permissive licence) · get_code("fcb15f76a1133e67")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 5 are also present in other papers whose code we hold (5 with at least one paper dated earlier, 0 the same month, 0 later) and 2 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
quantize
brick of this paper's assembly: helper
9 other papers
8 dated earlier, 1 undated · 1 only as an assembly brick
2210.17323 (2022-10)
GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers
2023.emnlp-main.892 (2023 (year only))
Z-FOLD: A Frustratingly Easy Post-Training Quantization Scheme for LLMs
2312.05215 (2023-12)
DeltaZip: Efficient Serving of Multiple Full-Model-Tuned LLMs
and 6 more not named here
quantize_activation_per_token_absmax
brick of this paper's assembly: helper
2 other papers
2 dated earlier
2405.12591 (2024-05)
Unlocking Data-free Low-bit Quantization with Matrix Decomposition for KV Cache Compression
2505.02214 (2025-05)
An Empirical Study of Qwen3 Quantization
quantize_weight_per_tensor_absmax
brick of this paper's assembly: helper
2 other papers
2 dated earlier
2405.12591 (2024-05)
Unlocking Data-free Low-bit Quantization with Matrix Decomposition for KV Cache Compression
2505.02214 (2025-05)
An Empirical Study of Qwen3 Quantization
Quantizer
brick of this paper's assembly: entry
1 other paper
1 dated earlier · 1 only as an assembly brick
2210.17323 (2022-10)
GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers
quantize_weight_per_channel_absmax
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2505.02214 (2025-05)
An Empirical Study of Qwen3 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 2026-11-21.

Repositories linked to this paper

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

Abstract

Compressing large language models (LLMs) for deployment on commodity GPUs remains challenging: conventional scalar quantization is limited to fixed bit-widths (e.g., 8/4/3-bit), offers only a few discrete compression points, and typically requires calibration data. We present FASQ (Flexible Accelerated Subspace Quantization), a calibration-free framework that applies product quantization to LLM weight matrices. By tuning two parameters, sub-vector size and codebook cardinality, FASQ exposes a continuous design space spanning 27-49% of the original FP16 model size, filling compression gaps that fixed-bit schemes cannot reach. On Meta-Llama-3-8B, FASQ surpasses 4-bit GPTQ and AWQ in accuracy (67.1-67.7 avg.) at 37-42% model size, with consistent results on Qwen3-8B and Qwen3.5-9B-Base. To make product quantization practical at inference time, we design custom CUDA kernels: a LUTfree direct-compute GEMV for decode and an output-stationary double-buffered LUT GEMM for prefill, both with split-K parallelism. On an RTX 3090, FASQ achieves 45.2 tok/s decode at effective 4-bit (2.56× memory reduction) and 51.8 tok/s at effective 3-bit (2.80×), both surpassing FP16 tensor-core performance (43.9 tok/s) and delivering 1.6-1.8× the throughput of AWQ, 2.2-2.5× of GPTQ, and 4.3-5.0× of RTN. FASQ is the only compressed method that accelerates decode beyond FP16, offering calibration-free compression, continuous size-quality trade-offs, and real-time inference on a single consumer GPU.

For agents

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

get_harvested_code_for_paper("2605.04084")
get_code_for_paper("2605.04084")
have("2605.04084")

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

curl https://syntology.ai/api/ran/2605.04084.json

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

[![Syntology run record](https://syntology.ai/api/ran/2605.04084.svg)](https://syntology.ai/paper/2605.04084)

Connect an agent — have() is free.