SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2107.14035 · 2021

ProtoTransformer: A Meta-Learning Approach to Providing Student Feedback

arXiv · PDF · Open in the Atlas

Code that ran

7 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 6 have no contract verdict; 1 failure of our instrument; 0 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/2107.14035.json · badge: /api/ran/2107.14035.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
mhw32/prototransformer-public community (classifier label)
classifier basis not recorded on the link
7 ran · 0 no recorded run
FunctionStatusWhere it lives
build_equivalences Ran
ran on a synthesized input
mhw32/prototransformer-public/src/datasets/feedback.py
code served (permissive licence) · get_code("f2e7fb4c01d6ee3b")
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 dict. 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)

# Construct indices - a numpy array of indices
indices = np.array([0, 1, 2, 3])

# Construct rubrics - a 2D numpy array where rows can be compared
# Each row represents a rubric, and we find which indices have equivalent rubrics
rubrics = np.array([
    [1, 2, 3, 4],
    [5, 6, 7, 8],
    [1, 2, 3, 4],
    [9, 10, 11, 12]
])

ARGS = {
    "indices": indices,
    "rubrics": rubrics
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
build_many_equivalences Ran
ran on a synthesized input
mhw32/prototransformer-public/src/datasets/feedback.py
code served (permissive licence) · get_code("26c93f0afd519a05")
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 dict. 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
import json

torch.manual_seed(0)

# Build test data for build_many_equivalences
# tasks: array of task identifiers
# rubrics: array of dicts where keys are JSON strings and values are booleans/scores

# Create tasks array with 2 distinct tasks
tasks = np.array(['task_a', 'task_a', 'task_b', 'task_b'])

# Create rubrics array
# Each rubric is a dict with JSON-stringified keys
rubric_1 = {
    json.dumps({'abbrev_label': 'clarity'}): True,
    json.dumps({'abbrev_label': 'completeness'}): False
}

rubric_2 = {
    json.dumps({'abbrev_label': 'clarity'}): True,
    json.dumps({'abbrev_label': 'completeness'}): False
}

rubric_3 = {
    json.dumps({'abbrev_label': 'clarity'}): False,
    json.dumps({'abbrev_label': 'completeness'}): True
}

rubric_4 = {
    json.dumps({'abbrev_label': 'clarity'}): False,
    json.dumps({'abbrev_label': 'completeness'}): True
}

rubrics = np.array([rubric_1, rubric_2, rubric_3, rubric_4], dtype=object)

ARGS = {
    'tasks': tasks,
    'rubrics': rubrics
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
create_position_ids_from_input_ids Ran
ran; our contract draft was wrong, not the code
mhw32/prototransformer-public/src/models/monkeypatch.py
code served (permissive licence) · get_code("3976dab6c53b60b0")
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 input_ids tensor: batch_size=2, sequence_length=4
# Contains token IDs with some padding tokens
input_ids = torch.tensor([
    [1, 2, 3, 0],      # batch 1: tokens 1,2,3 and padding token 0
    [4, 5, 0, 0]       # batch 2: tokens 4,5 and padding tokens 0,0
])

# padding_idx is typically 0 in most tokenizers
padding_idx = 0

# past_key_values_length represents offset from previous sequences in generation
past_key_values_length = 0

ARGS = {
    "input_ids": input_ids,
    "padding_idx": padding_idx,
    "past_key_values_length": past_key_values_length
}
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_domains Ran
ran on a synthesized input
mhw32/prototransformer-public/src/datasets/amazon.py
code served (permissive licence) · get_code("eee82740fe209a0f")
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 2, 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 os
import tempfile

# Create temporary directory with test files
temp_dir = tempfile.mkdtemp()

# Create filtered_name file with domain names
filtered_file = os.path.join(temp_dir, 'filtered_domains.txt')
with open(filtered_file, 'w', encoding='utf-8') as f:
    f.write('domain_a\n')
    f.write('domain_b\n')
    f.write('domain_c\n')
    f.write('domain_d\n')

# Create target_name file with test domains
target_file = os.path.join(temp_dir, 'target_domains.txt')
with open(target_file, 'w', encoding='utf-8') as f:
    f.write('domain_c\n')
    f.write('domain_d\n')

ARGS = {
    'data_path': temp_dir,
    'filtered_name': 'filtered_domains.txt',
    'target_name': 'target_domains.txt'
}
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 tuple, ndim 2, dtype family float — 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_test_data Ran
ran on a synthesized input
mhw32/prototransformer-public/src/datasets/amazon.py
code served (permissive licence) · get_code("5d2535cee5f10524")
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 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 os
import re
import copy
import string
import tempfile

# Create temporary directory and test data files
temp_dir = tempfile.mkdtemp()

# Create sample data files
domains = ['electronics', 'books']
usages = ['train', 'dev', 'test']
types = ['t2', 't4', 't5']

for domain in domains:
    for usage in usages:
        for t in types:
            filename = f"{domain}.{t}.{usage}"
            filepath = os.path.join(temp_dir, filename)
            with open(filepath, 'w', encoding='utf-8') as f:
                # Write negative examples (ending with -1)
                f.write("this product is bad-1\n")
                f.write("terrible quality-1\n")
                # Write positive examples (ending with no -1, just last char)
                f.write("great product1\n")
                f.write("excellent quality1\n")

ARGS = {
    'data_path': temp_dir,
    'domains': ['electronics', 'books']
}
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 tuple, ndim 1, dtype family float — 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_train_data Ran
ran on a synthesized input
mhw32/prototransformer-public/src/datasets/amazon.py
code served (permissive licence) · get_code("b86bcfe377f853aa")
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 dict. 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 os
import re
import string
import tempfile
import shutil

# Create a temporary directory with test data files
temp_dir = tempfile.mkdtemp()

# Create test data files
domains = ['domain1', 'domain2']
types = ['t2', 't4', 't5']

for domain in domains:
    for t in types:
        filename = f'{domain}.{t}.train'
        filepath = os.path.join(temp_dir, filename)
        with open(filepath, 'w', encoding='utf-8') as f:
            # Write negative samples (ending with -1)
            f.write('this is a negative sample-1\n')
            f.write('another bad example-1\n')
            # Write positive samples (ending with 1)
            f.write('this is a positive sample1\n')
            f.write('another good example1\n')

ARGS = {
    'data_path': temp_dir,
    'domains': ['domain1', 'domain2']
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
make_rubrics_float Ran
ran on a synthesized input
mhw32/prototransformer-public/src/datasets/feedback.py
code served (permissive licence) · get_code("759155413ac913f5")
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 int, shape (4, 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 torch
import numpy as np
import pandas as pd

# Create a small numpy array with mixed valid values and NaN
# Shape: (4, 3) - 4 rows, 3 columns
rubrics = np.array([
    [1.0, 'A', 10],
    [2.0, 'B', np.nan],
    [1.0, 'A', 10],
    [np.nan, 'C', 20]
], dtype=object)

ARGS = {
    'rubrics': rubrics
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, torch — 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 int, shape (4, 3) — shape (4, 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

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, 1 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 1 the same month, 1 later) and 6 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
create_position_ids_from_input_ids 28 other papers
1 the same month, 27 later
2107.00910 (2021-07)
Learned Token Pruning for Transformers
2201.11732 (2022-01)
IGLUE: A Benchmark for Transfer Learning across Modalities, Tasks, and Languages
2202.07962 (2022-02)
Revisiting Parameter-Efficient Tuning: Are We Really There Yet?
and 25 more not named here

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

High-quality computer science education is limited by the difficulty of providing instructor feedback to students at scale. While this feedback could in principle be automated, supervised approaches to predicting the correct feedback are bottlenecked by the intractability of annotating large quantities of student code. In this paper, we instead frame the problem of providing feedback as few-shot classification, where a meta-learner adapts to give feedback to student code on a new programming question from just a few examples annotated by instructors. Because data for meta-training is limited, we propose a number of amendments to the typical few-shot learning framework, including task augmentation to create synthetic tasks, and additional side information to build stronger priors about each task. These additions are combined with a transformer architecture to embed discrete sequences (e.g. code) to a prototypical representation of a feedback class label. On a suite of few-shot natural language processing tasks, we match or outperform state-of-the-art performance. Then, on a collection of student solutions to exam questions from an introductory university course, we show that our approach reaches an average precision of 88% on unseen questions, surpassing the 82% precision of teaching assistants. Our approach was successfully deployed to deliver feedback to 16,000 student exam-solutions in a programming course offered by a tier 1 university. This is, to the best of our knowledge, the first successful deployment of a machine learning based feedback to open-ended student code.

For agents

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

get_harvested_code_for_paper("2107.14035")
get_code_for_paper("2107.14035")
have("2107.14035")

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

curl https://syntology.ai/api/ran/2107.14035.json

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

[![Syntology run record](https://syntology.ai/api/ran/2107.14035.svg)](https://syntology.ai/paper/2107.14035)

Connect an agent — have() is free.