SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2009.03300 · ICLR · 2021

Measuring Massive Multitask Language Understanding

U Berkeley, Dan Hendrycks, Collin Burns, Jacob Steinhardt, Steven Uchicago, Andy Zou, Mantas Mazeika

arXiv · PDF · Open in the Atlas

Code that ran

19 of 26 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, 0 violated it and 9 have no contract verdict; 9 failures of our instrument; 7 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 2026-11-21; 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/2009.03300.json · badge: /api/ran/2009.03300.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
sanjass/MultiTaskLearning community (classifier label)
classifier basis not recorded on the link
9 ran · 7 no recorded run
ollmer/mmlu community (classifier label)
classifier basis: plan_row
9 ran · 0 no recorded run
aryopg/mmlu-redux extension (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
FunctionStatusWhere it lives
softmax Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ollmer/mmlu/evaluate.py
code served (permissive licence) · get_code("19f01570b0a18e2b")
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 ndarray, ndim 1, dtype family float, shape (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 2026-09-04T22:44:30Z). 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 softmax function takes a single parameter 'x' which should be array-like
# It uses: max(x), np.exp(z), np.sum(numerator), and division operations
# So x needs to be a numpy array or compatible

x = np.array([1.0, 2.0, 3.0, 4.0])

ARGS = {
    "x": x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 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
check_contains_num Ran
ran on a synthesized input
sanjass/MultiTaskLearning/ex2_data_parsing/data_parse.py
code served (permissive licence) · get_code("3a55bded83f59394")
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 re

def check_contains_num(question):
    pattern = re.compile('\\d+,\\d+|\\d*\\.\\d+|\\d+')
    for token in question[0].split(' '):
        if pattern.fullmatch(token):
            return True
    return False

# The function expects 'question' which is indexed with [0], suggesting it's a list or tuple
# The [0] element should be a string that can be split by spaces
# The function checks if any token matches a number pattern

ARGS = {
    "question": ["5 apples and 3.14 oranges"]
}
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
check_numeric Ran
ran on a synthesized input
sanjass/MultiTaskLearning/ex2_data_parsing/data_parse.py
code served (permissive licence) · get_code("e52cf9ef3b07a19c")
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 torch

def check_numeric(question):
    for i in range(1, 5):
        if not question[i].replace(',', '').isnumeric():
            return False
    return True

# Create a question object that supports indexing
# The function accesses indices 1, 2, 3, 4 and calls .replace() on each
# So it needs to be indexable and return strings with a replace method

question = [
    "non-numeric",  # index 0 (not accessed)
    "123",          # index 1
    "456",          # index 2
    "789",          # index 3
    "1,000"         # index 4 (has comma, will be replaced)
]

ARGS = {
    "question": question
}
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 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
create_model_path Ran
ran; our contract draft was wrong, not the code
aryopg/mmlu-redux/sft/sft_full.py
pointer only (licence: CC-BY-4.0) · get_code("7d2a2984ad40114e")
How it ran

The body of this sample is withheld: CC-BY-4.0, as GitHub detects it on this paper's repository, aryopg/mmlu-redux, is not in the permissive set Syntology serves inline. 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. A run is recorded.

format_example Ran
ran; our contract draft was wrong, not the code
ollmer/mmlu/evaluate.py
code served (permissive licence) · get_code("cd763eaf1ac287e7")
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) 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 pandas as pd

# Create a small DataFrame with the structure expected by format_example
# The function expects:
# - Column 0: prompt text
# - Columns 1 to k: answer choices (k columns)
# - Column k+1: the correct answer
# With choices = ['A', 'B', 'C', 'D'], we have k=4 choices

data = {
    0: ["What is 2+2?"],
    1: ["3"],
    2: ["4"],
    3: ["5"],
    4: ["6"],
    5: ["B"]  # The answer is 'B' (which corresponds to choice index 1, value "4")
}

df = pd.DataFrame(data)

ARGS = {
    'df': df,
    'idx': 0,
    'include_answer': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
pandas — 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
format_exp Ran
ran on a synthesized input
sanjass/MultiTaskLearning/036questions/utils.py
code served (permissive licence) · get_code("0a261ea0770f9244")
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

torch.manual_seed(0)

# format_exp takes a single parameter 'x' which is used in:
# - comparison: x < 0
# - negation: -1 * x
# - string conversion: str(x)
# This works with numeric types (int, float, or numeric tensors)

# Test with a simple numeric value
x = -5

ARGS = {
    'x': x
}
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
format_num Ran
ran on a synthesized input
sanjass/MultiTaskLearning/036questions/utils.py
code served (permissive licence) · get_code("4839a2077dc13fca")
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 numpy as np

torch.manual_seed(0)

# The function format_num takes a single parameter x
# It checks if x < 0, and if so returns 'negative ' + str(-1 * x)
# Otherwise it returns str(x)
# x can be any numeric type that supports comparison with 0 and negation

# Test with a positive number
x_positive = 42

ARGS = {
    'x': x_positive
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, 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
format_subject Ran
ran; our contract draft was wrong, not the code
ollmer/mmlu/evaluate.py
code served (permissive licence) · get_code("6ab745408cb8648b")
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) 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 format_subject(subject):
    l = subject.split('_')
    s = ''
    for entry in l:
        s += ' ' + entry
    return s

ARGS = {
    'subject': 'machine_learning_model'
}
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 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
gen_prompt Ran
ran; our contract draft was wrong, not the code
ollmer/mmlu/evaluate_flan.py
code served (permissive licence) · get_code("3383488ea82b6c15")
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 pandas as pd
import numpy as np

# Create a small DataFrame that mimics the structure expected by gen_prompt
# The function expects: question in column 0, answer choices in columns 1 to k, answer in column k+1
data = {
    0: ['What is 2+2?', 'What is the capital of France?'],
    1: ['3', 'London'],
    2: ['4', 'Paris'],
    3: ['5', 'Berlin'],
    4: ['6', 'Madrid'],
    5: ['A', 'B']  # answers column
}

train_df = pd.DataFrame(data)

subject = 'basic_math'

ARGS = {
    'train_df': train_df,
    'subject': subject,
    'k': -1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas — 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
gen_quant_cells Ran
ran on a synthesized input
sanjass/MultiTaskLearning/ex2_data_parsing/data_parse.py
code served (permissive licence) · get_code("29d8c9d894d5cb2e")
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 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
import torch.nn as nn

torch.manual_seed(0)

# Analyzing the function:
# - It takes a parameter 'question'
# - It accesses question[0] (first element)
# - It calls .split(' ') on question[0] (so question[0] must be a string)
# - It iterates over the length of the split result
# - It builds a list of indices

# The function expects question to be indexable (like a list or tuple)
# where the first element is a string that can be split by spaces

ARGS = {
    "question": ["What is the meaning of life?"]
}
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 1, 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
get_adaptive_bins Ran
ran; our contract draft was wrong, not the code
ollmer/mmlu/calib_tools.py
code served (permissive licence) · get_code("432838d3c5f2ab75")
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 1, dtype family float, shape (2,) 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 predictions (array-like) and num_bins (int)
# predictions should be sortable and indexable, containing numeric values
# num_bins should be a positive integer

predictions = np.array([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.95])
num_bins = 3

ARGS = {
    "predictions": predictions,
    "num_bins": num_bins,
}
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 1, dtype family float, shape (2,) — shape (2,) 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_and_update Ran
ran on a synthesized input
sanjass/MultiTaskLearning/036questions/writedata.py
code served (permissive licence) · get_code("0ae544dd79bf6e53")
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
import torch.nn as nn

torch.manual_seed(0)

class MockModule(nn.Module):
    def __init__(self):
        super().__init__()
    
    def return_data(self, train_id, test_id):
        # Returns (train, test, answers) tuples
        train = [torch.randn(2, 4), torch.randn(2, 4)]
        test = [torch.randn(2, 4)]
        answers = [1, 0]
        return train, test, answers

mod = MockModule()
train_id = 0
test_id = 0
train_data = []
test_data = []
test_answers = []

ARGS = {
    'mod': mod,
    'train_id': train_id,
    'test_id': test_id,
    'train_data': train_data,
    'test_data': test_data,
    'test_answers': test_answers
}
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_encoder Ran
ran; our contract draft was wrong, not the code
ollmer/mmlu/crop.py
code served (permissive licence) · get_code("3c57d0ee2854879f")
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 Encoder. 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 json
import os
import tempfile
import shutil

# Create a temporary directory for the mock encoder files
temp_dir = tempfile.mkdtemp()
models_dir = os.path.join(temp_dir, 'models', 'gpt2')
os.makedirs(models_dir, exist_ok=True)

# Create mock encoder.json file
encoder_dict = {
    'hello': 0,
    'world': 1,
    ' ': 2,
    'test': 3,
    '!': 4,
}
with open(os.path.join(models_dir, 'encoder.json'), 'w') as f:
    json.dump(encoder_dict, f)

# Create mock vocab.bpe file
bpe_content = "#version: 0.2\n"
bpe_content += "h e\n"
bpe_content += "w o\n"
bpe_content += "l d\n"
bpe_content += "t s\n"
bpe_content += "a b\n"
with open(os.path.join(models_dir, 'vocab.bpe'), 'w', encoding='utf-8') as f:
    f.write(bpe_content)

# Change to temp directory so relative paths work
original_cwd = os.getcwd()
os.chdir(temp_dir)

ARGS = {
    'model_name': 'gpt2'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
regex, requests, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Encoder — a Encoder output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Encoder 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_pairs Ran
ran; our contract draft was wrong, not the code
ollmer/mmlu/crop.py
code served (permissive licence) · get_code("4f76173dfe4b0c11")
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 set. 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 numpy as np

# The function get_pairs takes a word parameter
# Looking at the function body:
# - word is indexed with [0] to get the first element
# - word is iterated over with word[1:] 
# - Elements are added to a set as tuples (prev_char, char)
# So word should be a sequence/tuple of symbols (strings or any hashable objects)

# Create a simple word as a tuple of symbols
word = ("h", "e", "l", "l", "o")

ARGS = {
    "word": word
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set 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_quant_cells Ran
ran on a synthesized input
sanjass/MultiTaskLearning/036questions/utils.py
code served (permissive licence) · get_code("0e51b6d0af943f39")
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 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

def get_quant_cells(question):
    return [i for i in range(len(question.split(' ')))]

ARGS = {
    "question": "what is the answer to everything"
}
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 1, 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
infix_to_prefix Ran
ran on a synthesized input
sanjass/MultiTaskLearning/exercise2/util.py
code served (permissive licence) · get_code("87d40f0c0a1ce3ac")
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 1, dtype family float; 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.

# Analyze the function: infix_to_prefix
# The function takes an 'expression' parameter which is iterated over in reverse
# It processes characters that can be:
# - Operands (numbers/variables): single characters that aren't operators or parentheses
# - Operators: '+', '-', '*', '/'
# - Parentheses: '(' and ')'
# The function returns a list

# A simple valid expression string would work as the argument
# Let's create a basic infix expression like "3+4*2"

ARGS = {
    "expression": "3+4*2"
}
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 list, ndim 1, dtype family float — 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
mean Ran
ran, honoured the contract we drafted
ollmer/mmlu/calib_tools.py
code served (permissive licence) · get_code("0af29c87aa4bd6a8")
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 float64, ndim 0, dtype family float; 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 numpy as np

# The function takes inputs and computes mean if non-empty
# inputs should be an array-like object that supports len() and np.mean()

# Test case 1: non-empty array
inputs_nonempty = np.array([1.0, 2.0, 3.0, 4.0, 5.0])

# Test case 2: empty array
inputs_empty = np.array([])

# We'll use the non-empty case as the primary test
ARGS = {
    "inputs": inputs_nonempty
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
one_hot_encode Ran
ran; our contract draft was wrong, not the code
ollmer/mmlu/calib_tools.py
code served (permissive licence) · get_code("77933604fbca56e7")
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 (6, 3) 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

# Construct a small labels array with a few distinct classes
labels = np.array([0, 1, 2, 1, 0, 2])

# Define arguments for the function
ARGS = {
    "labels": labels,
    "num_classes": 3
}
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 (6, 3) — shape (6, 3) 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
tokenize_and_separate_quants Ran
ran on a synthesized input
sanjass/MultiTaskLearning/exercise2/util.py
code served (permissive licence) · get_code("4e014ec57f85bafe")
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 numpy as np
from collections import Counter
import itertools
import re

# Create sample data matching the function's expectations
data = [
    {
        'processed_question': 'What is 5 plus 3.5 and 2,000 divided by 4',
        'expression': '5 + 3.5 + 2000 / 4'
    },
    {
        'processed_question': 'Calculate 10 times 2.5 minus 1,500',
        'expression': '10 * 2.5 - 1500 + 7'
    },
    {
        'processed_question': 'Find 100 and 50 plus 25',
    }
]

n_min_vocab = 2

ARGS = {
    'data': data,
    'n_min_vocab': n_min_vocab
}
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 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
return_data Not yet run
unverified (harvested, no recorded run)
sanjass/MultiTaskLearning/036questions/filtering.py
code served (permissive licence) · get_code("69e44b454bef1c76")
How it ran

no invocation recorded.

return_data Not yet run
unverified (harvested, no recorded run)
sanjass/MultiTaskLearning/036questions/filtering2.py
code served (permissive licence) · get_code("b441fdbef0a06567")
How it ran

no invocation recorded.

return_data Not yet run
unverified (harvested, no recorded run)
sanjass/MultiTaskLearning/036questions/gradientdescent.py
code served (permissive licence) · get_code("bddd1344d4501303")
How it ran

no invocation recorded.

return_data Not yet run
unverified (harvested, no recorded run)
sanjass/MultiTaskLearning/036questions/gradientdescent2.py
code served (permissive licence) · get_code("5dfa113ed2d4f658")
How it ran

no invocation recorded.

return_data Not yet run
unverified (harvested, no recorded run)
sanjass/MultiTaskLearning/036questions/logisticregression.py
code served (permissive licence) · get_code("de6faacbeb0eb06d")
How it ran

no invocation recorded.

return_data Not yet run
unverified (harvested, no recorded run)
sanjass/MultiTaskLearning/036questions/rnn.py
code served (permissive licence) · get_code("69930c08eceb6a88")
How it ran

no invocation recorded.

setup Not yet run
unverified (harvested, no recorded run)
sanjass/MultiTaskLearning/exercise2/util.py
code served (permissive licence) · get_code("26f49fbc83128412")
How it ran

no invocation recorded.

Also present in other papers' code

Of 26 samples lifted from this paper's linked repositories, 26 carry a function identity and 0 do not. Of the 26 distinct functions with an identity, 10 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 10 later) and 16 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
get_pairs 258 other papers
10 dated earlier, 247 later, 1 undated · 4 only as an assembly brick
1802.06474 (2018-02)
A Closed-form Solution to Photorealistic Image Stylization
1803.01128 (2018-03)
Seq2Sick: Evaluating the Robustness of Sequence-to-Sequence Models with Adversarial Examples
1809.00125 (2018-09)
Simple Fusion: Return of the Language Model
and 255 more not named here
format_subject 21 other papers
21 later
2304.12986 (2023-04)
Measuring Massive Multitask Chinese Understanding
2305.17306 (2023-05)
Chain-of-Thought Hub: A Continuous Effort to Measure Large Language Models' Reasoning Performance
2310.01651 (2023-10)
Fool Your (Vision and) Language Model with Embarrassingly Simple Permutations
and 18 more not named here
format_example 16 other papers
16 later
2305.17306 (2023-05)
Chain-of-Thought Hub: A Continuous Effort to Measure Large Language Models' Reasoning Performance
2310.05736 (2023-10)
LLMLingua: Compressing Prompts for Accelerated Inference of Large Language Models
2312.07104 (2023-12)
SGLang: Efficient Execution of Structured Language Model Programs
and 13 more not named here
gen_prompt 10 other papers
10 later
2305.17306 (2023-05)
Chain-of-Thought Hub: A Continuous Effort to Measure Large Language Models' Reasoning Performance
2310.05736 (2023-10)
LLMLingua: Compressing Prompts for Accelerated Inference of Large Language Models
2312.00700 (2023-12)
WeGeFT: Weight-Generative Fine-Tuning for Multi-Faceted Efficient Adaptation of Large Models
and 7 more not named here
softmax 7 other papers
7 later
2310.04928 (2023-10)
Large Language Models Only Pass Primary School Exams in Indonesia: A Comprehensive Test on IndoMMLU
2310.09550 (2023-10)
Can Large Language Model Comprehend Ancient Chinese? A Preliminary Test on ACLUE
2402.12840 (2024-02)
ArabicMMLU: Assessing Massive Multitask Language Understanding in Arabic
and 4 more not named here
one_hot_encode 4 other papers
3 later, 1 undated
2401.02718 (2024-01)
Calibration Attacks: A Comprehensive Study of Adversarial Attacks on Model Confidence
2403.01244 (2024-03)
Mitigating Catastrophic Forgetting in Large Language Models with Self-Synthesized Rehearsal
2601.16503 (2026-01)
MRAG: Benchmarking Retrieval-Augmented Generation for Bio-medicine
and 1 more not named here
get_adaptive_bins 3 other papers
2 later, 1 undated
2403.01244 (2024-03)
Mitigating Catastrophic Forgetting in Large Language Models with Self-Synthesized Rehearsal
2601.16503 (2026-01)
MRAG: Benchmarking Retrieval-Augmented Generation for Bio-medicine
openreview_g9G7qyAzki
Generalized Correctness Models: Learning Calibrated and Cross-Model Correctness Predictors from Historical Patterns
mean 3 other papers
2 later, 1 undated
2403.01244 (2024-03)
Mitigating Catastrophic Forgetting in Large Language Models with Self-Synthesized Rehearsal
2601.16503 (2026-01)
MRAG: Benchmarking Retrieval-Augmented Generation for Bio-medicine
openreview_g9G7qyAzki
Generalized Correctness Models: Learning Calibrated and Cross-Model Correctness Predictors from Historical Patterns
get_encoder 2 other papers
2 later
2403.01244 (2024-03)
Mitigating Catastrophic Forgetting in Large Language Models with Self-Synthesized Rehearsal
2601.16503 (2026-01)
MRAG: Benchmarking Retrieval-Augmented Generation for Bio-medicine
create_model_path 1 other paper
1 later
2406.04127 (2024-06)
Are We Done with MMLU?

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

We propose a new test to measure a text model's multitask accuracy. The test covers 57 tasks including elementary mathematics, US history, computer science, law, and more. To attain high accuracy on this test, models must possess extensive world knowledge and problem solving ability. We find that while most recent models have near random-chance accuracy, the very largest GPT-3 model improves over random chance by almost 20 percentage points on average. However, on every one of the 57 tasks, the best models still need substantial improvements before they can reach expert-level accuracy. Models also have lopsided performance and frequently do not know when they are wrong. Worse, they still have nearrandom accuracy on some socially important subjects such as morality and law. By comprehensively evaluating the breadth and depth of a model's academic and professional understanding, our test can be used to analyze models across many tasks and to identify important shortcomings.

For agents

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

get_harvested_code_for_paper("2009.03300")
get_code_for_paper("2009.03300")
have("2009.03300")

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

curl https://syntology.ai/api/ran/2009.03300.json

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

[![Syntology run record](https://syntology.ai/api/ran/2009.03300.svg)](https://syntology.ai/paper/2009.03300)

Connect an agent — have() is free.