SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2405.14852 · 2024

PV-Tuning: Beyond Straight-Through Estimation for Extreme LLM Compression

arXiv · PDF · Open in the Atlas

Code that ran

5 of 10 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; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 3 failures of our instrument; 5 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/2405.14852.json · badge: /api/ran/2405.14852.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
vahe1994/aqlm canonical (classifier label)
classifier basis: census
5 ran · 5 no recorded run
FunctionStatusWhere it lives
get_int_dtype Ran
ran; our contract draft was wrong, not the code
vahe1994/aqlm/inference_lib/src/aqlm/utils.py
code served (permissive licence) · get_code("286ca2263c661514")
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 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)

ARGS = {
    "nbits": 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 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
get_num_layers Ran
ran, honoured the contract we drafted
vahe1994/aqlm/convert_to_hf.py
code served (permissive licence) · get_code("18920618b7b5ee81")
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 int, ndim 0, dtype family int, 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
import torch.nn as nn


class ConfigStandIn:
    """Minimal config class matching function requirements"""
    def __init__(self, model_type, num_hidden_layers=None):
        self.model_type = model_type
        if num_hidden_layers is not None:
            self.num_hidden_layers = num_hidden_layers


# Test with a valid model type
config_llama = ConfigStandIn(model_type='llama', num_hidden_layers=12)

ARGS = {
    'config': config_llama
}
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 int, ndim 0, dtype family int, 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
next_power_of_2 Ran
ran, honoured the contract we drafted
vahe1994/aqlm/inference_lib/src/aqlm/inference_kernels/triton_kernel.py
code served (permissive licence) · get_code("9782bd07382e951c")
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 int, ndim 0, dtype family int, 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 math

def next_power_of_2(x):
    return 1 if x == 0 else 2 ** math.ceil(math.log2(x))

# Test with various values
ARGS = {
    "x": 5
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, 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
pack_int_data Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
vahe1994/aqlm/inference_lib/src/aqlm/utils.py
code served (permissive licence) · get_code("0e1845ecac45aea3")
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 (5,) 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 integer tensor with values that will test the nbits logic
# We need values that may be >= 2**(nbits-1) to trigger the subtraction
data = torch.tensor([0, 1, 127, 128, 255], dtype=torch.int64)

# nbits parameter - let's use 8 to test the int8 dtype path
nbits = 8

ARGS = {
    "data": data,
    "nbits": nbits,
}
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 (5,) — shape (5,) 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
unpack_int_data Ran
ran; our contract draft was wrong, not the code
vahe1994/aqlm/inference_lib/src/aqlm/utils.py
code served (permissive licence) · get_code("a1a49cbe29b10116")
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, 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 IntTensor with values that can be unpacked
data = torch.randint(0, 256, (2, 4), dtype=torch.int32)

# nbits parameter - number of bits to extract
nbits = 8

ARGS = {
    "data": data,
    "nbits": nbits,
}
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, 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
get_backward_pass_kernel Not yet run
unverified (harvested, no recorded run)
vahe1994/aqlm/inference_lib/src/aqlm/inference_kernels/kernel_selector.py
code served (permissive licence) · get_code("8443c635b944859f")
How it ran

no invocation recorded.

get_forward_pass_kernel Not yet run
unverified (harvested, no recorded run)
vahe1994/aqlm/inference_lib/src/aqlm/inference_kernels/kernel_selector.py
code served (permissive licence) · get_code("4538301eff106753")
How it ran

no invocation recorded.

numba_gemm_lut Not yet run
unverified (harvested, no recorded run)
vahe1994/aqlm/inference_lib/src/aqlm/inference_kernels/numba_kernel.py
code served (permissive licence) · get_code("db8f26f618d4290e")
How it ran

no invocation recorded.

rgetattr Not yet run
unverified (harvested, no recorded run)
vahe1994/aqlm/convert_legacy_model_format.py
code served (permissive licence) · get_code("44925cf3831bfe79")
How it ran

no invocation recorded.

rsetattr Not yet run
unverified (harvested, no recorded run)
vahe1994/aqlm/convert_legacy_model_format.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from DBO-DKFZ/uncertainty-benchmark, is pointer only (licence: NONE) · get_code("f9de4cdd8a624801")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from DBO-DKFZ/uncertainty-benchmark, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 distinct functions with an identity, 3 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 3 later) and 7 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
rgetattr 13 other papers
11 dated earlier, 2 later · 1 only as an assembly brick
2203.04036 (2022-03)
StyleHEAT: One-Shot High-Resolution Editable Talking Face Generation via Pre-trained StyleGAN
2209.02952 (2022-09)
BiFuse++: Self-supervised and Efficient Bi-projection Fusion for 360 Depth Estimation
2211.05778 (2022-11)
InternImage: Exploring Large-Scale Vision Foundation Models with Deformable Convolutions
and 10 more not named here
rsetattr 8 other papers
6 dated earlier, 2 later
2203.04036 (2022-03)
StyleHEAT: One-Shot High-Resolution Editable Talking Face Generation via Pre-trained StyleGAN
2212.04501 (2022-12)
Learning Video Representations from Large Language Models
2305.13417 (2023-05)
VISIT: Visualizing and Interpreting the Semantic Information Flow of Transformers
and 5 more not named here
next_power_of_2 1 other paper
1 later
2505.13430 (2025-05)
Fine-tuning Quantized Neural Networks with Zeroth-order Optimization

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

There has been significant interest in "extreme" compression of large language models (LLMs), i.e., to 1-2 bits per parameter, which allows such models to be executed efficiently on resource-constrained devices. Existing work focused on improved one-shot quantization techniques and weight representations; yet, purely post-training approaches are reaching diminishing returns in terms of the accuracy-vs-bit-width trade-off. State-of-the-art quantization methods such as QuIP# and AQLM include fine-tuning (part of) the compressed parameters over a limited amount of calibration data; however, such fine-tuning techniques over compressed weights often make exclusive use of straight-through estimators (STE), whose performance is not well-understood in this setting. In this work, we question the use of STE for extreme LLM compression, showing that it can be sub-optimal, and perform a systematic study of quantization-aware fine-tuning strategies for LLMs. We propose PV-Tuning - a representation-agnostic framework that generalizes and improves upon existing fine-tuning strategies, and provides convergence guarantees in restricted cases. On the practical side, when used for 1-2 bit vector quantization, PV-Tuning outperforms prior techniques for highly-performant models such as Llama and Mistral. Using PV-Tuning, we achieve the first Pareto-optimal quantization for Llama 2 family models at 2 bits per parameter.

For agents

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

get_harvested_code_for_paper("2405.14852")
get_code_for_paper("2405.14852")
have("2405.14852")

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

curl https://syntology.ai/api/ran/2405.14852.json

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

[![Syntology run record](https://syntology.ai/api/ran/2405.14852.svg)](https://syntology.ai/paper/2405.14852)

Connect an agent — have() is free.