SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.17700 · ACL · 2024

RAVEL: Evaluating Interpretability Methods on Disentangling Language Model Representations

Jing Huang, Christopher Potts, Atticus Geiger, Zhengxuan Wu, Mor Geva

arXiv · PDF · Open in the Atlas

Code that ran

12 of 17 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 12 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/2402.17700.json · badge: /api/ran/2402.17700.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
explanare/ravel canonical (classifier label)
classifier basis: census
12 ran · 5 no recorded run
FunctionStatusWhere it lives
compute_cross_entropy_loss Ran
ran; our contract draft was wrong, not the code
explanare/ravel/src/utils/metric_utils.py
code served (permissive licence) · get_code("39d29ba2dce082a9")
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 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
import torch.nn as nn

torch.manual_seed(0)

# Small tensors: batch_size=2, seq_len=5, vocab_size=4
batch_size = 2
seq_len = 5
vocab_size = 4
next_n_tokens = 1

# Create logits tensor: shape (batch_size, seq_len, vocab_size)
logits = torch.randn(batch_size, seq_len, vocab_size)

# Create labels tensor: shape (batch_size, seq_len)
# Use values in range [0, vocab_size) and include pad_token_id
labels = torch.randint(0, vocab_size, (batch_size, seq_len))

# Set pad_token_id
pad_token_id = 0

ARGS = {
    "logits": logits,
    "labels": labels,
    "pad_token_id": pad_token_id,
    "next_n_tokens": next_n_tokens
}
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 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
compute_disentangle_score Ran
ran; our contract draft was wrong, not the code
explanare/ravel/src/utils/metric_utils.py
code served (permissive licence) · get_code("618cb3bd9cfb4a14")
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 numpy as np

# Construct log_data dictionary
# The function accesses log_data[t]['metrics']['base_labels']['accuracy'] 
# and log_data[t]['metrics']['labels']['accuracy']
log_data = {
    'task_1': {
        'metrics': {
            'base_labels': {'accuracy': 0.85},
            'labels': {'accuracy': 0.90}
        }
    },
    'task_2': {
        'metrics': {
            'base_labels': {'accuracy': 0.80},
            'labels': {'accuracy': 0.88}
        }
    },
    'task_3': {
        'metrics': {
            'base_labels': {'accuracy': 0.92},
            'labels': {'accuracy': 0.95}
        }
    },
    'task_4': {
        'metrics': {
            'base_labels': {'accuracy': 0.78},
            'labels': {'accuracy': 0.82}
        }
    }
}

# Construct attribute_to_iso_tasks dictionary
# Maps attributes to lists of task names
attribute_to_iso_tasks = {
    'color': ['task_1', 'task_2'],
    'shape': ['task_3', 'task_4']
}

# Construct attribute_to_cause_tasks dictionary
# Maps attributes to lists of task names
attribute_to_cause_tasks = {
    'rotation': ['task_1', 'task_3'],
    'scale': ['task_2', 'task_4']
}

ARGS = {
    'log_data': log_data,
    'attribute_to_iso_tasks': attribute_to_iso_tasks,
    'attribute_to_cause_tasks': attribute_to_cause_tasks
}
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 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
compute_metrics Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
explanare/ravel/src/utils/metric_utils.py
code served (permissive licence) · get_code("5500053238475cfe")
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

torch.manual_seed(0)

# Construct eval_preds: list of tensors with shape (batch_size, seq_len, vocab_size)
# Using 3D tensors (logits before argmax)
eval_preds = [
    torch.randn(2, 5, 10),  # batch=2, seq_len=5, vocab_size=10
    torch.randn(2, 5, 10),
]

# Construct eval_labels: list of tensors with shape (batch_size, seq_len)
# Values should be token ids (0-9) or pad_token_id (which is 0) or negative (ignored)
eval_labels = [
    torch.tensor([[1, 2, 3, 4, 5], [2, 3, 4, 5, 6]]),  # batch=2, seq_len=5
    torch.tensor([[0, 1, 2, 3, 4], [5, 6, 7, 8, 0]]),  # batch=2, seq_len=5
]

pad_token_id = 0

ARGS = {
    'eval_preds': eval_preds,
    'eval_labels': eval_labels,
    'pad_token_id': pad_token_id,
    'last_n_tokens': 1,
}
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 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
compute_principal_component Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
explanare/ravel/src/methods/pca.py
code served (permissive licence) · get_code("ebcb0699723a970c")
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 numpy as np
from sklearn.decomposition import PCA

# Set random seed for reproducibility
np.random.seed(0)

# Create small feature matrix: 2 samples, 4 features
features = np.random.randn(2, 4).astype(np.float32)

# Set n_components to be less than min(n_samples, n_features)
n_components = 2

ARGS = {
    'features': features,
    'n_components': n_components
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn — 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
generate_batched Ran
ran; our contract draft was wrong, not the code
explanare/ravel/src/utils/generation_utils.py
code served (permissive licence) · get_code("e7618272c2024815")
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 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 torch
import torch.nn as nn

torch.manual_seed(0)

class MockTokenizer:
    def __init__(self):
        self.pad_token_id = 0
    
    def __call__(self, text, return_tensors=None, padding=None, max_length=None, truncation=None):
        # Simulate tokenization
        if isinstance(text, str):
            token_ids = [1, 2, 3]
        else:
            token_ids = [[1, 2, 3] for _ in text]
        
        if max_length:
            token_ids = [ids[:max_length] + [self.pad_token_id] * max(0, max_length - len(ids)) 
                        for ids in token_ids] if isinstance(token_ids[0], list) else token_ids[:max_length] + [self.pad_token_id] * max(0, max_length - len(token_ids))
        
        if isinstance(token_ids[0], list):
            input_ids = torch.tensor(token_ids, dtype=torch.long)
            attention_mask = torch.tensor([[1 if x != self.pad_token_id else 0 for x in ids] for ids in token_ids], dtype=torch.long)
        else:
            input_ids = torch.tensor([token_ids], dtype=torch.long)
            attention_mask = torch.tensor([[1 if x != self.pad_token_id else 0 for x in token_ids]], dtype=torch.long)
        
        if return_tensors == 'pt':
            return {
                'input_ids': input_ids,
                'attention_mask': attention_mask
            }
        return {'input_ids': input_ids}
    
    def batch_decode(self, token_ids, skip_special_tokens=False):
        # Mock decode: return a string for each sequence
        batch_size = token_ids.shape[0]
        return [f"generated text {i}" for i in range(batch_size)]

class MockGenerativeModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 4)
        self.device_type = 'cpu'
    
    @property
    def device(self):
        return torch.device('cpu')
    
    def generate(self, input_ids, attention_mask=None, max_new_tokens=None, 
                 do_sample=False, num_return_sequences=1, return_dict_in_generate=False, 
                 pad_token_id=None, **kwargs):
        # Generate mock output token ids
        batch_size = input_ids.shape[0]
        seq_len = input_ids.shape[1] + (max_new_tokens if max_new_tokens else 5)
        output_ids = torch.randint(1, 100, (batch_size * num_return_sequences, seq_len), dtype=torch.long)
        return output_ids

tokenizer = MockTokenizer()
model = MockGenerativeModel()

all_prompts = [
    "What is machine learning?",
    "Explain neural networks.",
]

ARGS = {
    'pretrained_model': model,
    'tokenizer': tokenizer,
    'all_prompts': all_prompts,
    'max_length': 64,
    'prompt_max_length': 32,
    'max_new_tokens': 32,
    'sample_n': 2,
    'batch_size': 2,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 2, 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
get_score Ran
ran; our contract draft was wrong, not the code
explanare/ravel/src/methods/linear_adversarial_probe.py
code served (permissive licence) · get_code("4637704289ff6a8a")
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 numpy as np
import sklearn
from sklearn.linear_model import SGDClassifier
from sklearn.metrics import log_loss

# Set random seed for reproducibility
np.random.seed(0)

# Create small synthetic datasets
# X_train: (samples=5, features=4)
X_train = np.random.randn(5, 4).astype(np.float32)
y_train = np.array([0, 1, 0, 1, 0])

# X_dev: (samples=3, features=4)
X_dev = np.random.randn(3, 4).astype(np.float32)
y_dev = np.array([0, 1, 0])

# P: positive semi-definite matrix (4x4) for projection
# Create a symmetric positive definite matrix
A = np.random.randn(4, 4).astype(np.float32)
P = A @ A.T + np.eye(4)

# rank: smaller than feature dimension
rank = 2

ARGS = {
    'X_train': X_train,
    'y_train': y_train,
    'X_dev': X_dev,
    'y_dev': y_dev,
    'P': P,
    'rank': rank
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn — 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
init_sgd_classifier Ran
ran; our contract draft was wrong, not the code
explanare/ravel/src/methods/linear_adversarial_probe.py
code served (permissive licence) · get_code("9a0c2776f5ac2bc0")
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 SGDClassifier. 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.

from sklearn.linear_model import SGDClassifier

def init_sgd_classifier(params=None):
    if params is None:
        params = {'loss': 'log_loss', 'tol': 0.0001, 'iters_no_change': 15, 'alpha': 0.0001, 'max_iter': 25000}
    return SGDClassifier(loss=params['loss'], fit_intercept=True, max_iter=params['max_iter'], tol=params['tol'], n_iter_no_change=params['iters_no_change'], n_jobs=32, alpha=params['alpha'])

ARGS = {
    'params': {
        'loss': 'log_loss',
        'tol': 0.0001,
        'iters_no_change': 15,
        'alpha': 0.0001,
        'max_iter': 25000
    }
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
sklearn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type SGDClassifier — a SGDClassifier output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a SGDClassifier 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
kept_first_n_label_token Ran
ran; our contract draft was wrong, not the code
explanare/ravel/src/utils/dataset_utils.py
code served (permissive licence) · get_code("f648fddc170ec6d8")
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

torch.manual_seed(0)

x = {
    'labels': torch.tensor([[1, 2, 3, 4, 5], [6, 7, 8, 9, 10]]),
    'inv_labels': torch.tensor([[10, 20, 30, 40, 50], [60, 70, 80, 90, 100]]),
}

ARGS = {
    'x': x,
    'first_n': 2,
    'padding_offset': 3,
}
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 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
kept_first_n_label_token_multitask Ran
ran; our contract draft was wrong, not the code
explanare/ravel/src/utils/dataset_utils.py
code served (permissive licence) · get_code("f404f9aad2c1b27e")
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

torch.manual_seed(0)

# Construct the arguments for kept_first_n_label_token_multitask
x = {
    'labels': torch.tensor([1, 2, 3, 4, 5, 6, 7, 8]),
    'inv_labels': torch.tensor([10, 20, 30, 40, 50, 60, 70, 80]),
    'split': 'train',
    'base_labels': None,
}

cause_tasks = ['train', 'val']
first_n = 4
padding_offset = 3

ARGS = {
    'x': x,
    'cause_tasks': cause_tasks,
    'first_n': first_n,
    'padding_offset': padding_offset,
}
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 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
preproc_tokenize Ran
ran; our contract draft was wrong, not the code
explanare/ravel/src/utils/dataset_utils.py
code served (permissive licence) · get_code("39872fcdf40c8524")
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 copy
import torch
from transformers import AutoTokenizer

torch.manual_seed(0)

class MockTokenizer:
    def __init__(self):
        self.padding_side = 'left'
        self.pad_token_id = 0
    
    def __call__(self, text, padding=None, max_length=None, return_tensors=None, truncation=None):
        if isinstance(text, list):
            batch_size = len(text)
        else:
            batch_size = 1
        
        # Return a dict with input_ids and attention_mask
        return {
            'input_ids': torch.randint(1, 100, (batch_size, max_length or 10)),
            'attention_mask': torch.ones(batch_size, max_length or 10, dtype=torch.long)
        }

tokenizer = MockTokenizer()
tokenizer.padding_side = 'left'

examples = {
    'input': ['hello world', 'foo bar'],
    'label': ['world hello', 'bar foo'],
    'extra_input_field': ['extra text one', 'extra text two'],
    'extra_label_field': ['label extra one', 'label extra two']
}

max_input_length = 10
max_output_length = 8
input_feature = 'input'
label_feature = 'label'
extra_input_to_tokenize = ['extra_input_field']
extra_label_to_tokenize = ['extra_label_field']

ARGS = {
    'tokenizer': tokenizer,
    'max_input_length': max_input_length,
    'max_output_length': max_output_length,
    'examples': examples,
    'input_feature': input_feature,
    'label_feature': label_feature,
    'extra_input_to_tokenize': extra_input_to_tokenize,
    'extra_label_to_tokenize': extra_label_to_tokenize
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — 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
remove_invalid_token_id Ran
ran; our contract draft was wrong, not the code
explanare/ravel/src/utils/intervention_utils.py
code served (permissive licence) · get_code("fe8f9421a66f408f")
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 tensor with some -100 values (invalid token ids)
token_ids = torch.tensor([
    [1, -100, 3, 4],
    [5, 6, -100, 8]
], dtype=torch.long)

pad_id = 2

ARGS = {
    "token_ids": token_ids,
    "pad_id": pad_id
}
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
select_features_with_classifier Ran
ran; our contract draft was wrong, not the code
explanare/ravel/src/methods/select_features.py
code served (permissive licence) · get_code("be76384e0b33672a")
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 torch.nn as nn
import numpy as np
from sklearn.svm import LinearSVC
from sklearn.feature_selection import SelectFromModel

torch.manual_seed(0)

# Create a simple featurizer that transforms inputs to a feature space
class SimpleFeaturizer(nn.Module):
    def __init__(self, input_dim, output_dim):
        super().__init__()
        self.linear = nn.Linear(input_dim, output_dim)
    
    def forward(self, x):
        return self.linear(x)

# Build minimal valid arguments
featurizer = SimpleFeaturizer(input_dim=3, output_dim=8)

# Create small batch of inputs: shape (2, 3) - batch of 2, 3 features
inputs = torch.randn(2, 3)

# Create labels for binary classification: shape (2,)
labels = np.array([0, 1])

# coeff will use default if not provided, but we can also pass None explicitly
coeff = None

ARGS = {
    "featurizer": featurizer,
    "inputs": inputs,
    "labels": labels,
    "coeff": coeff
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn, 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
gen_context_test_split Not yet run
unverified (harvested, no recorded run)
explanare/ravel/src/utils/generate_ravel_instance.py
code served (permissive licence) · get_code("8240c646692f8698")
How it ran

no invocation recorded.

gen_entity_test_split Not yet run
unverified (harvested, no recorded run)
explanare/ravel/src/utils/generate_ravel_instance.py
code served (permissive licence) · get_code("58779741197dded2")
How it ran

no invocation recorded.

gen_train_split Not yet run
unverified (harvested, no recorded run)
explanare/ravel/src/utils/generate_ravel_instance.py
code served (permissive licence) · get_code("8289a1e4195d682d")
How it ran

no invocation recorded.

symmetric Not yet run
unverified (harvested, no recorded run)
explanare/ravel/src/methods/linear_adversarial_probe.py
pointer only (licence: MIT); the stored copy get_code serves, from shauli-ravfogel/rlace-icml, is pointer only (licence: NONE) · get_code("5eb6f7201a4d2f1c")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from shauli-ravfogel/rlace-icml, 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.

train_intervention_step Not yet run
unverified (harvested, no recorded run)
explanare/ravel/src/utils/intervention_utils.py
code served (permissive licence) · get_code("01750c9c0ff161d4")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 2 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 1 later) and 15 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
generate_batched 1 other paper
1 later
2407.17817 (2024-07)
Demystifying Verbatim Memorization in Large Language Models
symmetric 1 other paper
1 dated earlier
2201.12091 (2022-01)
Linear Adversarial Concept Erasure

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

Individual neurons participate in the representation of multiple high-level concepts. To what extent can different interpretability methods successfully disentangle these roles? To help address this question, we introduce RAVEL (Resolving Attribute-Value Entanglements in Language Models), a dataset that enables tightly controlled, quantitative comparisons between a variety of existing interpretability methods. We use the resulting conceptual framework to define the new method of Multi-task Distributed Alignment Search (MDAS), which allows us to find distributed representations satisfying multiple causal criteria. With Llama2-7B as the target language model, MDAS achieves state-of-the-art results on RAVEL, demonstrating the importance of going beyond neuron-level analyses to identify features distributed across activations. We release our benchmark at https://github.com/ explanare/ravel.

For agents

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

get_harvested_code_for_paper("2402.17700")
get_code_for_paper("2402.17700")
have("2402.17700")

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

curl https://syntology.ai/api/ran/2402.17700.json

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

[![Syntology run record](https://syntology.ai/api/ran/2402.17700.svg)](https://syntology.ai/paper/2402.17700)

Connect an agent — have() is free.