SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1911.02549 · 2019

MLPerf Inference Benchmark

arXiv · PDF · Open in the Atlas

Code that ran

18 of 21 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 17 have no contract verdict; 1 failure of our instrument; 3 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/1911.02549.json · badge: /api/ran/1911.02549.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
mlcommons/inference community (classifier label)
classifier basis not recorded on the link
18 ran · 3 no recorded run
FunctionStatusWhere it lives
apply_argmax Ran
ran on a synthesized input
mlcommons/inference/vision/medical_imaging/3d-unet-kits19/inference_utils.py
code served (permissive licence) · get_code("3867e52e00cf507d")
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 4, dtype family int, shape (1, 1, 4, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create a small test image with shape (bs=1, channel=3, height=4, width=4)
# This represents a sub-volume inference result with 3 channels
test_image = np.random.randn(1, 3, 4, 4).astype(np.float32)

ARGS = {
    "image": test_image
}
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 4, dtype family int, shape (1, 1, 4, 4) — shape (1, 1, 4, 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
apply_norm_map Ran
ran on a synthesized input
mlcommons/inference/vision/medical_imaging/3d-unet-kits19/inference_utils.py
code served (permissive licence) · get_code("3cea92bceb1a2c9b")
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 4, dtype family float, shape (2, 4, 3, 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

torch.manual_seed(0)

# Create small tensors for image and norm_map
# The function performs element-wise division: image /= norm_map
# Both need to be broadcastable and have compatible shapes

image = torch.randn(2, 4, 3, 3)  # batch=2, channels=4, height=3, width=3
norm_map = torch.randn(2, 4, 3, 3) + 1.0  # same shape, offset to avoid near-zero values

ARGS = {
    "image": image,
    "norm_map": norm_map,
}
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 Tensor, ndim 4, dtype family float, shape (2, 4, 3, 3) — shape (2, 4, 3, 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
calculate_retrieval_metrics Ran
ran on a synthesized input
mlcommons/inference/e2e-rag/accuracy_eval.py
code served (permissive licence) · get_code("49b7d9bb406a744e")
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
from typing import Dict, List

torch.manual_seed(0)

# Setup for calculate_retrieval_metrics
# The function takes two parameters:
# 1. retrieved_urls: List[str] - a list of retrieved URL strings
# 2. expected_urls: List[str] - a list of expected/ground-truth URL strings

# Create realistic but minimal test data
retrieved_urls = [
    "https://example.com/page1",
    "https://example.com/page2",
    "https://example.com/page3",
    "https://example.com/page4"
]

expected_urls = [
    "https://example.com/page2",
    "https://example.com/page3",
    "https://example.com/page5"
]

ARGS = {
    "retrieved_urls": retrieved_urls,
    "expected_urls": expected_urls
}
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 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
calculate_retrieval_metrics Ran
ran on a synthesized input
mlcommons/inference/e2e-rag/evaluation.py
code served (permissive licence) · get_code("ad3ffba7eb3bd285")
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 arguments for calculate_retrieval_metrics
expected_urls = [
    "https://example.com/page1",
    "https://example.com/page2",
    "https://example.com/page3",
]

retrieved_urls = [
    "https://example.com/page2",
    "https://example.com/page4",
    "https://example.com/page1",
    "https://example.com/page5",
    "https://example.com/page3",
]

k_values = [1, 3, 5, 10]

ARGS = {
    "expected_urls": expected_urls,
    "retrieved_urls": retrieved_urls,
    "k_values": k_values,
}
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 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
call_judge Ran
ran on a synthesized input
mlcommons/inference/e2e-rag/accuracy_eval.py
code served (permissive licence) · get_code("7a7d20420db85d84")
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 json
import os
import re
from typing import Dict, List

# Set up mock environment
os.environ['OPENROUTER_API_KEY'] = 'sk-or-v1-test-key-12345'

# Define arguments for call_judge function
ARGS = {
    'question': 'What is the capital of France?',
    'ground_truth': 'Paris',
    'llm_answer': 'Paris',
    'service_url': 'http://127.0.0.1:8123/v1/chat/completions',
    'model_name': 'gpt-oss-20b',
    'api_key': 'sk-or-v1-test-key-12345'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
requests — 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
call_judge Ran
ran on a synthesized input
mlcommons/inference/e2e-rag/evaluate.py
code served (permissive licence) · get_code("1ce3b34b77863636")
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 requests
from unittest.mock import Mock, MagicMock
import json

# Create a mock requests.Session
session = Mock(spec=requests.Session)

# Create a mock response object
mock_response = Mock()
mock_response.json.return_value = {
    'choices': [
        {
            'message': {
                'content': json.dumps({
                    'score': 1,
                    'explanation': 'The model answer correctly answers the question.'
                })
            }
        }
    ]
}
mock_response.raise_for_status.return_value = None

# Configure the session mock to return our mock response
session.post.return_value = mock_response

# Define the arguments
ARGS = {
    'session': session,
    'service_url': 'https://api.openrouter.ai/api/v1/chat/completions',
    'model': 'gpt-3.5-turbo',
    'question': 'What is the capital of France?',
    'gold': 'Paris',
    'pred': 'Paris'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
requests — 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
compute_md5 Ran
ran on a synthesized input
mlcommons/inference/e2e-rag/datasetup_accuracy_eval.py
code served (permissive licence) · get_code("007cfc74e0e0649b")
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 tempfile
import os

# Create a temporary file with some content
temp_file = tempfile.NamedTemporaryFile(delete=False, mode='wb')
temp_file.write(b"Hello, this is test content for MD5 hashing.\n" * 100)
temp_file.close()

ARGS = {
    "file_path": temp_file.name
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
evaluate_results Ran
ran on a synthesized input
mlcommons/inference/e2e-rag/accuracy_eval.py
code served (permissive licence) · get_code("54a637337a4e3cb3")
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 pandas as pd
import tempfile
import os

# Create a minimal TSV dataset file
tsv_content = """Prompt	Answer	wikipedia_link_1	wikipedia_link_2
What is the capital of France?	Paris	https://en.wikipedia.org/wiki/Paris	https://en.wikipedia.org/wiki/France
What is the largest planet?	Jupiter	https://en.wikipedia.org/wiki/Jupiter	https://en.wikipedia.org/wiki/Solar_System
"""

# Write to a temporary file
temp_dir = tempfile.mkdtemp()
dataset_path = os.path.join(temp_dir, 'frames_dataset.tsv')
with open(dataset_path, 'w') as f:
    f.write(tsv_content)

# Create results dict mapping query_id to result_dict
# The function expects results to have queries that match the dataset
results = {
    'query_1': {
        'query': 'What is the capital of France?',
        'llm_answer': 'Paris',
        'retrieved_urls': ['https://en.wikipedia.org/wiki/Paris', 'https://en.wikipedia.org/wiki/France']
    },
    'query_2': {
        'query': 'What is the largest planet?',
        'llm_answer': 'Jupiter is the largest planet in our solar system',
        'retrieved_urls': ['https://en.wikipedia.org/wiki/Jupiter']
    }
}

ARGS = {
    'results': results,
    'dataset_path': dataset_path,
    'num_workers': 2,
    'judge_service_url': 'http://127.0.0.1:8123/v1/chat/completions',
    'judge_model': 'gpt-oss-20b'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas, requests — 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
evaluate_retrieval_query Ran
ran on a synthesized input
mlcommons/inference/e2e-rag/evaluation.py
code served (permissive licence) · get_code("305c35ca9e02614f")
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
from typing import List, Dict, Any, Optional, Tuple, Union, Callable


class MockDocument:
    """Minimal Document class matching what the function expects"""
    def __init__(self, page_content: str, metadata: Dict[str, Any]):
        self.page_content = page_content
        self.metadata = metadata


class MockRerankerModel(nn.Module):
    """Minimal reranker model"""
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(10, 1)
    
    def forward(self, x):
        return self.linear(x)


class MockRAGDB:
    """Minimal RAG database matching function expectations"""
    def __init__(self, reranker=True):
        self._reranker_model = MockRerankerModel() if reranker else None
    
    def lookup(self, query: str, k: int = 50) -> List[MockDocument]:
        """Return mock retrieval results"""
        results = [
            MockDocument(
                page_content=f"Content for document {i}",
                metadata={"original_url": f"https://example.com/doc{i}"}
            )
            for i in range(min(k, 5))
        ]
        return results
    
    def rerank(self, query: str, passages: List[str]) -> List[Tuple[str, float]]:
        """Return reranked passages with scores"""
        scored = [(passage, 0.9 - i * 0.1) for i, passage in enumerate(passages)]
        return sorted(scored, key=lambda x: x[1], reverse=True)


# Build arguments for evaluate_retrieval_query
torch.manual_seed(0)

rag_db = MockRAGDB(reranker=True)

query = "What is machine learning?"

expected_urls = [
    "https://example.com/doc0",
    "https://example.com/doc1",
    "https://example.com/doc3",
]

ARGS = {
    "rag_db": rag_db,
    "query": query,
    "expected_urls": expected_urls,
    "top_k_retriever": 50,
    "top_k_reranking": 10,
    "verbose": True,
    "no_rerank": False,
    "retrieval_strategy": "fixed_k",
    "print_results": False,
    "return_results": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
retrieve, 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
extract_wikipedia_links Ran
ran on a synthesized input
mlcommons/inference/e2e-rag/download_docs.py
code served (permissive licence) · get_code("757290e048b0c991")
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.

import re
import ast

# Analyze the function to understand what it expects:
# - extract_wikipedia_links takes one parameter: `item`
# - It checks if item is a string, and if so, uses ast.literal_eval to convert it
# - Then it iterates over `item` as a list
# - For each entry in the list, it finds Wikipedia URLs using regex
# - The regex pattern matches Wikipedia URLs in the format: https://en.wikipedia.org/wiki/...
# - URLs can contain commas (legitimately), and the function handles Markdown link syntax

# Construct a valid `item` argument:
# Option 1: A list of strings containing Wikipedia URLs
# Option 2: A string representation of a list (which gets literal_eval'd)

# Let's create test data with both markdown links and plain URLs with commas
item_value = [
    "Check out https://en.wikipedia.org/wiki/Quincy,_Massachusetts and https://en.wikipedia.org/wiki/John_Doe",
    "[Wikipedia Link](https://en.wikipedia.org/wiki/Python_(programming_language))",
    "https://en.wikipedia.org/wiki/Tick,_Tick..._Boom!",
]

ARGS = {
    "item": item_value
}
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
fix_malformed_url Ran
ran on a synthesized input
mlcommons/inference/e2e-rag/download_docs.py
code served (permissive licence) · get_code("4ceeb0dd1f0e41e8")
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 re

def fix_malformed_url(url: str) -> str:
    if '#:~:text=' in url:
        url = url.split('#:~:text=')[0]
    if re.search('%25[0-9A-Fa-f]{2}', url):
        url = re.sub('%25([0-9A-Fa-f]{2})', lambda m: '%' + m.group(1), url)
    open_parens = url.count('(')
    close_parens = url.count(')')
    if open_parens > close_parens:
        missing_closes = open_parens - close_parens
        url += ')' * missing_closes
    return url

ARGS = {
    "url": "https://example.com/page(test#:~:text=hello%25world"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_config Ran
ran; our contract draft was wrong, not the code
mlcommons/inference/text_to_video/wan-2.2-t2v-a14b/run_inference.py
code served (permissive licence) · get_code("44c6d1dcec31f489")
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 yaml
import tempfile
import os

# Create a temporary YAML configuration file
temp_dir = tempfile.mkdtemp()
config_file = os.path.join(temp_dir, "config.yaml")

config_content = """
model:
  type: transformer
  hidden_size: 256
  num_layers: 2
training:
  learning_rate: 0.001
  batch_size: 32
  epochs: 10
data:
  path: /path/to/data
  split: 0.8
"""

with open(config_file, 'w') as f:
    f.write(config_content)

ARGS = {
    "config_path": config_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
yaml — 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
load_prompts Ran
ran on a synthesized input
mlcommons/inference/text_to_video/wan-2.2-t2v-a14b/run_inference.py
code served (permissive licence) · get_code("7b3e0f2610f104fb")
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.

import tempfile
import os

# Create a temporary file with sample prompts
temp_dir = tempfile.mkdtemp()
dataset_path = os.path.join(temp_dir, "prompts.txt")

with open(dataset_path, 'w') as f:
    f.write("What is machine learning?\n")
    f.write("Explain neural networks\n")
    f.write("\n")
    f.write("How does backpropagation work?\n")

ARGS = {
    "dataset_path": dataset_path
}
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
load_results Ran
ran on a synthesized input
mlcommons/inference/e2e-rag/evaluate.py
code served (permissive licence) · get_code("a9ede96a0650926d")
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 tempfile
import json
import pickle
from pathlib import Path
import pandas as pd

# Create a temporary directory for test files
tmpdir = tempfile.mkdtemp()

# Setup for pickle file path
pkl_path = Path(tmpdir) / "results.pkl"
df = pd.DataFrame([
    {'query': 'What is 2+2?', 'llm_answer': '4', 'success': True},
    {'query': 'What is the capital of France?', 'llm_answer': 'Paris', 'success': True},
    {'query': 'Bad query', 'llm_answer': 'Error', 'success': False},
])
with open(pkl_path, 'wb') as f:
    pickle.dump(df, f)

# Setup for JSON file path
json_path = Path(tmpdir) / "results.json"
json_data = {
    'results': [
        {'prompt': 'What is AI?', 'llm_answer': 'Artificial Intelligence'},
        {'prompt': 'What is ML?', 'llm_answer': 'Machine Learning'},
        {'prompt': None, 'llm_answer': 'No prompt'},
    ]
}
json_path.write_text(json.dumps(json_data), encoding='utf-8')

# Define ARGS for both test cases
ARGS = {
    'path': pkl_path  # Can also test with json_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — 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
parse_accuracy_log Ran
ran on a synthesized input
mlcommons/inference/e2e-rag/datasetup_accuracy_eval.py
code served (permissive licence) · get_code("97cde6f95685bb1a")
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 json
import tempfile
import os

# Create a temporary JSON file with valid accuracy log data
temp_dir = tempfile.mkdtemp()
log_file = os.path.join(temp_dir, "mlperf_log_accuracy.json")

# Create sample accuracy log in JSON array format
sample_data = [
    {
        "qsl_idx": 0,
        "data": "48656c6c6f",
        "seq_id": 1
    },
    {
        "qsl_idx": 1,
        "data": "576f726c64",
        "seq_id": 2
    },
    {
        "qsl_idx": 2,
        "data": [65, 66, 67],
        "seq_id": 3
    }
]

with open(log_file, 'w') as f:
    json.dump(sample_data, f)

ARGS = {
    "log_path": log_file
}
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
postprocess Ran
ran on a synthesized input
mlcommons/inference/lon/network_SUT.py
code served (permissive licence) · get_code("f5c0874dcb5d50cd")
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)

# The function takes a single parameter 'query' and returns it unchanged
# Based on the function body, 'query' can be any object
# We'll use a simple dict as a representative query object

query_obj = {
    "text": "sample query",
    "id": 123,
    "data": torch.randn(2, 4)
}

ARGS = {
    "query": query_obj
}
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 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
preprocess Ran
ran on a synthesized input
mlcommons/inference/lon/network_SUT.py
code served (permissive licence) · get_code("e05733aa3409d7b2")
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 preprocess takes a single parameter 'query'
# and simply returns it unchanged. We need to provide a valid argument.
# Since the function doesn't specify the type, we can pass various types.
# A common use case would be passing a string, dict, tensor, or list.
# Let's use a simple string as the query.

query = "test query string"

ARGS = {
    "query": query
}
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
validate_database Ran
ran on a synthesized input
mlcommons/inference/e2e-rag/datasetup_accuracy_eval.py
code served (permissive licence) · get_code("7d14cc5656b8c82f")
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
from pathlib import Path
import tempfile
import os

torch.manual_seed(0)

# Create a minimal stand-in for VectorDB with the required attributes and methods
class MockFaissIndex:
    def __init__(self, d=768):
        self.d = d
        self.ntotal = 5

class MockVectorStore:
    def __init__(self):
        self.index = MockFaissIndex(d=768)
        self.index_to_docstore_id = {0: "doc0", 1: "doc1", 2: "doc2", 3: "doc3", 4: "doc4"}
        self.docstore = {
            "doc0": "Document 0 content",
            "doc1": "Document 1 content",
            "doc2": "Document 2 content",
            "doc3": "Document 3 content",
            "doc4": "Document 4 content"
        }

class MockVectorDB:
    def __init__(self, retriever_model=None, device='cpu', embedding_device='cpu', load_embeddings=False):
        self.retriever_model = retriever_model
        self.device = device
        self.embedding_device = embedding_device
        self._vector_store = MockVectorStore()
    
    def from_serialized(self, database_path):
        pass
    
    def lookup(self, query, k=1):
        # Return k mock results
        return [
            {"id": 0, "content": "result 0", "score": 0.95},
            {"id": 1, "content": "result 1", "score": 0.87},
            {"id": 2, "content": "result 2", "score": 0.79},
        ][:k]

# Patch the retrieve module import
import sys
from unittest.mock import MagicMock

retrieve_module = MagicMock()
retrieve_module.VectorDB = MockVectorDB
sys.modules['retrieve'] = retrieve_module

# Create temporary database path
with tempfile.TemporaryDirectory() as tmpdir:
    database_path = os.path.join(tmpdir, "test_database.db")
    
    # Create a dummy file to represent the database
    with open(database_path, 'w') as f:
        f.write("mock database content")
    
    retriever_model_path = "mock_retriever_model"

ARGS = {
    'database_path': database_path if os.path.exists(database_path) else str(Path(tempfile.gettempdir()) / "test_database.db"),
    'retriever_model': retriever_model_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, retrieve, 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
dnn_model Not yet run
unverified (harvested, no recorded run)
mlcommons/inference/lon/network_SUT.py
code served (permissive licence) · get_code("2d9995b9375a7f1f")
How it ran

no invocation recorded.

download_frames_dataset Not yet run
unverified (harvested, no recorded run)
mlcommons/inference/e2e-rag/download_docs.py
code served (permissive licence) · get_code("d9fde455e8c45d1d")
How it ran

no invocation recorded.

gaussian_kernel Not yet run
unverified (harvested, no recorded run)
mlcommons/inference/vision/medical_imaging/3d-unet-kits19/inference_utils.py
code served (permissive licence) · get_code("dd5f36b69e453200")
How it ran

no invocation recorded.

Also present in other papers' code

Of 21 samples lifted from this paper's linked repositories, 21 carry a function identity and 0 do not. Of the 21 distinct functions with an identity, 1 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 1 later) and 20 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
load_config 3 other papers
3 later
2311.12085 (2023-11)
Pyramid Diffusion for Fine 3D Large Scene Generation
2311.17034 (2023-11)
Telling Left from Right: Identifying Geometry-Aware Semantic Correspondence
2403.16379 (2024-03)
FlashEval: Towards Fast and Accurate Evaluation of Text-to-image Diffusion Generative Models

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

Machine-learning (ML) hardware and software system demand is burgeoning. Driven by ML applications, the number of different ML inference systems has exploded. Over 100 organizations are building ML inference chips, and the systems that incorporate existing models span at least three orders of magnitude in power consumption and five orders of magnitude in performance; they range from embedded devices to data-center solutions. Fueling the hardware are a dozen or more software frameworks and libraries. The myriad combinations of ML hardware and ML software make assessing ML-system performance in an architecture-neutral, representative, and reproducible manner challenging. There is a clear need for industry-wide standard ML benchmarking and evaluation criteria. MLPerf Inference answers that call. In this paper, we present our benchmarking method for evaluating ML inference systems. Driven by more than 30 organizations as well as more than 200 ML engineers and practitioners, MLPerf prescribes a set of rules and best practices to ensure comparability across systems with wildly differing architectures. The first call for submissions garnered more than 600 reproducible inference-performance measurements from 14 organizations, representing over 30 systems that showcase a wide range of capabilities. The submissions attest to the benchmark's flexibility and adaptability.

For agents

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

get_harvested_code_for_paper("1911.02549")
get_code_for_paper("1911.02549")
have("1911.02549")

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

curl https://syntology.ai/api/ran/1911.02549.json

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

[![Syntology run record](https://syntology.ai/api/ran/1911.02549.svg)](https://syntology.ai/paper/1911.02549)

Connect an agent — have() is free.