SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2203.06462 · ACL · 2022

Low-Rank Softmax Can Have Unargmaxable Classes in Theory but Rarely in Practice

Nikolay Bogoychev, Adam Lopez, Andreas Grivas

arXiv · PDF · Open in the Atlas

Code that ran

6 of 9 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 6 have no contract verdict; 0 failures of our instrument; 3 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/2203.06462.json · badge: /api/ran/2203.06462.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
andreasgrv/unargmaxable canonical (classifier label)
classifier basis not recorded on the link
6 ran · 3 no recorded run
FunctionStatusWhere it lives
datetime_aware Ran
ran on a synthesized input
andreasgrv/unargmaxable/stollen/server/utils.py
code served (permissive licence) · get_code("45de5c654cc93d67")
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 datetime

# The function datetime_aware takes a single parameter 'o'
# It checks if o is an instance of datetime.date or datetime.datetime
# If so, it calls strftime('%c') on it
# Otherwise it returns o as-is

# Create test cases for both branches
date_obj = datetime.date(2024, 1, 15)
datetime_obj = datetime.datetime(2024, 1, 15, 14, 30, 45)
non_datetime_obj = "regular string"

# The function takes one parameter 'o', so we need to provide a value for it
# Let's test with a datetime.datetime object
ARGS = {
    'o': datetime_obj
}
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 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
filter_modelname Ran
ran on a synthesized input
andreasgrv/unargmaxable/paper/plots/plot_row_iterations.py
code served (permissive licence) · get_code("86b19aa45d183d79")
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

def filter_modelname(name):
    name = name.replace('bigx2', 'big')
    name = name.replace(':model.npz', '')
    name = name.replace('.npz', '')
    name = name.replace(':model', '')
    return name

ARGS = {
    'name': 'bigx2_model:model.npz'
}
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 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
get_perp_vecs Ran
ran on a synthesized input
andreasgrv/unargmaxable/paper/plots/stolen_probability.py
code served (permissive licence) · get_code("1f0fabc8eaea5e8e")
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 ndarray, ndim 2, dtype family float, shape (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 numpy as np

# The function takes a single parameter 'vecs' which should be a numpy array
# Looking at the function body:
# - vecs is copied
# - It's indexed with [:, 0] and [:, 1] so it needs at least 2 columns
# - It's used in dot product with vecs.T
# - It's normalized with np.linalg.norm along axis=1
# So vecs should be a 2D numpy array with shape (n_samples, n_features) where n_features >= 2

# Create a small valid input array
vecs = np.array([
    [2.0, 3.0, 1.0, 0.5],
    [1.5, 2.5, 0.8, 0.3],
    [3.0, 4.0, 1.2, 0.7]
], dtype=np.float64)

ARGS = {
    "vecs": vecs
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (3, 4) — shape (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
is_valid_token Ran
ran on a synthesized input
andreasgrv/unargmaxable/stollen/utils.py
code served (permissive licence) · get_code("4e5ea1d24adc191d")
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 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 unicodedata as ud

lang_map = {'el': 'GREEK', 'en': 'LATIN', 'de': 'LATIN', 'ru': 'CYRILLIC', 'he': 'HEBREW', 'ar': 'ARABIC'}

def token_cleanup(token):
    token = token.strip()
    if token.startswith('▁'):
        token = token[1:]
    if token.endswith('</w>'):
        token = token[:-4]
    return token

def is_valid_token(lang, token):
    assert ' ' not in token, '"%s"' % token
    token = token_cleanup(token)
    valid = False
    try:
        c_langs = [ud.name(c).split(' ', 1)[0] for c in token]
        is_lang = all((l == lang_map[lang] for l in c_langs))
        valid = is_lang
    except Exception as e:
        print(token, e)
        valid = False
    return valid

ARGS = {
    'lang': 'en',
    'token': 'hello'
}
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 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
softmax Ran
ran on a synthesized input
andreasgrv/unargmaxable/paper/plots/stolen_probability.py
code served (permissive licence) · get_code("a612b0389a399641")
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 ndarray, 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 numpy as np

# Create a small batch of logits with shape (2, 4)
# - batch size: 2
# - number of classes: 4
inputs = np.array([
    [1.0, 2.0, 3.0, 4.0],
    [0.5, 1.5, 2.5, 3.5]
], dtype=np.float32)

ARGS = {
    "inputs": inputs
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, 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
token_cleanup Ran
ran on a synthesized input
andreasgrv/unargmaxable/stollen/utils.py
code served (permissive licence) · get_code("0ccc3c33c0c30035")
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

ARGS = {
    "token": "▁hello</w>"
}
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 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
N Not yet run
unverified (harvested, no recorded run)
andreasgrv/unargmaxable/paper/appendix/braid-slice-regions/count_regions.py
code served (permissive licence) · get_code("a144bf26251239be")
How it ran

no invocation recorded.

S Not yet run
unverified (harvested, no recorded run)
andreasgrv/unargmaxable/paper/appendix/braid-slice-regions/count_regions.py
code served (permissive licence) · get_code("f7c6b59a740e3542")
How it ran

no invocation recorded.

T Not yet run
unverified (harvested, no recorded run)
andreasgrv/unargmaxable/paper/appendix/braid-slice-regions/count_regions.py
code served (permissive licence) · get_code("6b4db4e8842f98c8")
How it ran

no invocation recorded.

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 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 9 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

Classifiers in natural language processing (NLP) often have a large number of output classes. For example, neural language models (LMs) and machine translation (MT) models both predict tokens from a vocabulary of thousands. The Softmax output layer of these models typically receives as input a dense feature representation, which has much lower dimensionality than the output. In theory, the result is some words may be impossible to be predicted via argmax, irrespective of input features, and empirically, there is evidence this happens in small language models (Demeter et al., 2020). In this paper we ask whether it can happen in practical large language models and translation models. To do so, we develop algorithms to detect such unargmaxable tokens in public models. We find that 13 out of 150 models do indeed have such tokens; however, they are very infrequent and unlikely to impact model quality. We release our algorithms and code so that others can test their models. 1

For agents

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

get_harvested_code_for_paper("2203.06462")
get_code_for_paper("2203.06462")
have("2203.06462")

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

curl https://syntology.ai/api/ran/2203.06462.json

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

[![Syntology run record](https://syntology.ai/api/ran/2203.06462.svg)](https://syntology.ai/paper/2203.06462)

Connect an agent — have() is free.