ARIA/mcq-bench
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Initial commit: MCQ benchmark with checkpoint/resume support

Browse Source
This commit is contained in:
ARIA
2026-07-14 14:32:24 +02:00
commit fd40133142
5 changed files with 704 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.gitignore
+4
View File
@@ -0,0 +1,4 @@
runs/
__pycache__/
*.pyc
config.yaml
LICENSE
+176
View File
@@ -0,0 +1,176 @@
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to the Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademark. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work, You may choose to offer, and charge a fee for, acceptance
of support, warranty, indemnity, or other liability obligations
and/or rights consistent with this License. However, in accepting
such obligations, You may act only on Your own behalf and on Your
own responsibility, not on behalf of any other Contributor, and
only if You agree to indemnify, defend, and hold each Contributor
harmless for any liability incurred by, or claims asserted against,
such Contributor by reason of your accepting any such warranty or
additional liability.
END OF TERMS AND CONDITIONS
README.md
+54
View File
@@ -0,0 +1,54 @@
# MCQ Benchmark
LLM benchmark for evaluating models on multiple-choice questions from a parquet dataset (financial reasoning).
## Features
- **Checkpoint/Resume** — progress is saved after each question. Stop and resume anytime.
- **Live Status** — real-time progress with running accuracy percentage.
- **Configurable** — YAML config for model endpoint, parameters, and benchmark settings.
## Setup
1. Copy the example config:
```bash
cp config.example.yaml config.yaml
```
2. Edit `config.yaml` with your model endpoint and parameters.
## Usage
```bash
# Run benchmark (prompts to resume if checkpoint exists)
python3 benchmark.py
# Auto-resume from latest checkpoint
python3 benchmark.py --resume
# Start fresh, ignoring existing checkpoint
python3 benchmark.py --fresh
# Use custom config
python3 benchmark.py --config my_config.yaml
```
## Output
Results are saved to `runs/<model_name>_YYYYMMDD_HHMMSS.json` with full details:
- Overall accuracy and timing
- Per-question results (correct, incorrect, timeouts)
- Model configuration used
Checkpoints (`*.checkpoint.json`) are stored in `runs/` and auto-deleted on completion.
## Live Output Example
```
[3048/5074] what is AAPL expecting in return for Q4 2016 based on this sheet:... CORRECT [67.4%]
[3049/5074] calculate the total revenue for the fiscal year 2015... WRONG (expected: B, got: A) [67.4%]
```
## License
Apache 2.0
benchmark.py
+439
View File
@@ -0,0 +1,439 @@
#!/usr/bin/env python3
"""
MCQ Benchmark
Evaluates local LLMs on multiple-choice questions from a parquet dataset.
Usage:
python3 benchmark.py [--config config.yaml] [--resume] [--fresh]
Checkpoints are saved after each question to runs/<model>_<timestamp>.checkpoint.json.
Use --resume to continue from the latest checkpoint.
"""
import argparse
import json
import re
import sys
import time
from datetime import datetime
from pathlib import Path
import pandas as pd
import requests
import yaml
def load_config(path: str) -> dict:
with open(path, "r") as f:
return yaml.safe_load(f)
def sanitize_model_name(name: str) -> str:
sanitized = re.sub(r'[^\w\s-]', '', name)
sanitized = re.sub(r'[\s]+', '_', sanitized)
return sanitized.strip('_') or 'unknown_model'
def extract_choice(response: str) -> str:
"""Extract the chosen option letter from the model response."""
if not response:
return ""
cleaned = response.strip()
# Check for letter in parentheses: (A), (B), etc.
match = re.search(r'\(([A-Z])\)', cleaned)
if match:
return match.group(1).upper()
# "Answer: A", "Choice: B", etc. (colon separator)
match = re.search(r'(?:answer|choice|option)\s*:\s*["\']?([A-Z])["\']?', cleaned, re.IGNORECASE)
if match:
return match.group(1).upper()
# "Answer is A", "Choice is B", etc.
match = re.search(r'(?:answer|choice|option)\s+is\s+["\']?([A-Z])["\']?', cleaned, re.IGNORECASE)
if match:
return match.group(1).upper()
# "A)", "B)", etc.
match = re.search(r'\b([A-Z])\)', cleaned)
if match:
return match.group(1).upper()
# Single letter response
match = re.match(r'^\s*([A-Z])\s*$', cleaned)
if match:
return match.group(1).upper()
return ""
def answers_match(expected_key: str, model_response: str, options: dict) -> bool:
"""Compare the expected answer key with the model's response."""
if not expected_key:
return False
expected = expected_key.strip().upper()
# Direct match on letter
choice = extract_choice(model_response)
if choice == expected:
return True
# Check if the model's response contains the correct option text
correct_text = options.get(expected, "")
if correct_text:
model_lower = model_response.lower()
correct_lower = correct_text.lower()
# Remove "million", "billion" suffixes for numeric comparison
clean_correct = re.sub(r'\s*(million|billion|thousand|percent|%)\b', '', correct_lower).strip()
if clean_correct and clean_correct in model_lower:
return True
return False
def build_prompt(row) -> str:
"""Build the prompt from a dataset row."""
query = row.get('query', '')
options = row.get('options', {})
# Extract the question text from the query
question_match = re.search(r'Question:\s*(.+?)(?:\nAnswer:|$)', query, re.DOTALL)
question = question_match.group(1).strip() if question_match else ""
# Extract the context from the query
context_match = re.search(r'Context:\s*(.+?)(?:\nQuestion:|$)', query, re.DOTALL)
context = context_match.group(1).strip() if context_match else ""
# Format options
options_text = ""
if isinstance(options, dict):
for key, value in sorted(options.items()):
options_text += f" ({key}) {value}\n"
elif isinstance(options, str):
try:
parsed = json.loads(options)
if 'options' in parsed:
for key, value in sorted(parsed['options'].items()):
options_text += f" ({key}) {value}\n"
except (json.JSONDecodeError, TypeError):
pass
prompt = f"""You are a financial analyst answering multiple-choice questions based on the provided context.
**Context:**
{context}
**Question:** {question}
**Options:**
{options_text}
Answer with ONLY the letter of the correct option (e.g., "A", "B", "C", or "D"). Do not include any explanation."""
return prompt
def query_model(prompt: str, config: dict) -> str | None:
"""Send a request to the LLM endpoint and return the response text."""
model_cfg = config['model']
endpoint = f"{model_cfg['endpoint']}/chat/completions"
headers = {
"Content-Type": "application/json",
"Authorization": f"Bearer {model_cfg['api_key']}",
}
payload = {
"model": model_cfg['name'],
"messages": [
{"role": "system", "content": "You are a precise financial analyst. Answer multiple-choice questions based on the provided context. Respond with only the letter of the correct option."},
{"role": "user", "content": prompt},
],
"max_tokens": model_cfg['max_tokens'],
"temperature": model_cfg.get('temperature', 0.0),
}
try:
resp = requests.post(
endpoint,
headers=headers,
json=payload,
timeout=model_cfg['timeout'],
)
resp.raise_for_status()
data = resp.json()
return data['choices'][0]['message']['content']
except requests.exceptions.Timeout:
return None
except Exception as e:
print(f" Warning: Error - {e}", file=sys.stderr)
return ""
def find_latest_checkpoint(output_dir: Path, safe_name: str) -> tuple:
"""Find the latest checkpoint file for a model and return (path, checkpoint_data)."""
checkpoints = sorted(output_dir.glob(f"{safe_name}_*.checkpoint.json"))
if checkpoints:
with open(checkpoints[-1], 'r') as f:
return checkpoints[-1], json.load(f)
return None, None
def save_checkpoint(checkpoint_path: Path, results: dict, completed_ids: list, start_time: float):
"""Save checkpoint with current progress."""
checkpoint = {
'completed_ids': completed_ids,
'results': results,
'start_time': start_time,
'saved_at': datetime.now().isoformat(),
}
with open(checkpoint_path, 'w') as f:
json.dump(checkpoint, f, indent=2, ensure_ascii=False)
def build_report(config: dict, results: dict, start_time: float):
"""Build the final report from results."""
model_cfg = config['model']
elapsed = time.time() - start_time
total_attempted = len(results['correct']) + len(results['incorrect']) + len(results['timeouts'])
accuracy = (len(results['correct']) / total_attempted * 100) if total_attempted > 0 else 0
return {
'benchmark': 'MCQ Benchmark (parquet)',
'timestamp': datetime.now().isoformat(),
'model': {
'name': model_cfg['name'],
'endpoint': model_cfg['endpoint'],
'timeout': model_cfg['timeout'],
'max_tokens': model_cfg['max_tokens'],
'temperature': model_cfg.get('temperature', 0.0),
},
'summary': {
'total_questions': total_attempted,
'correct': len(results['correct']),
'incorrect': len(results['incorrect']),
'timeouts': len(results['timeouts']),
'accuracy_pct': round(accuracy, 1),
'elapsed_seconds': round(elapsed, 1),
'avg_seconds_per_question': round(elapsed / total_attempted, 2) if total_attempted > 0 else 0,
},
'correct_answers': results['correct'],
'incorrect_answers': results['incorrect'],
'timeout_questions': results['timeouts'],
}
def print_summary(report: dict, output_path: Path):
"""Print the final summary."""
s = report['summary']
print(f"\n{'='*60}")
print(f" RESULTS")
print(f"{'='*60}")
print(f" Accuracy: {s['accuracy_pct']:.1f}% ({s['correct']}/{s['total_questions']})")
print(f" Correct: {s['correct']}")
print(f" Wrong: {s['incorrect']}")
print(f" Timeouts: {s['timeouts']}")
if s['total_questions'] > 0:
print(f" Time: {s['elapsed_seconds']:.1f}s ({s['avg_seconds_per_question']:.1f}s/q)")
if report['incorrect_answers']:
print(f"\n Wrong Answers ({len(report['incorrect_answers'])}):")
for item in report['incorrect_answers'][:10]:
print(f" [{item['id']}] {item['question'][:70]}...")
print(f" Expected: {item['expected_key']}")
print(f" Got: {item['model_response'][:70]}")
if len(report['incorrect_answers']) > 10:
print(f" ... and {len(report['incorrect_answers']) - 10} more")
print(f"\n Report saved: {output_path}")
print(f"{'='*60}\n")
def run_benchmark(config_path: str, resume: bool = False, fresh: bool = False):
config = load_config(config_path)
model_cfg = config['model']
bench_cfg = config['benchmark']
output_dir = Path(config_path).parent / bench_cfg.get('output_dir', 'runs')
output_dir.mkdir(exist_ok=True)
safe_name = sanitize_model_name(model_cfg['name'])
# Check for existing checkpoint
checkpoint_path, checkpoint_data = find_latest_checkpoint(output_dir, safe_name)
has_checkpoint = checkpoint_data is not None
if has_checkpoint and not fresh:
if resume or input(f" Found checkpoint: {checkpoint_path.name}\n {len(checkpoint_data['completed_ids'])} questions completed. Resume? [y/N]: ").strip().lower() == 'y':
print(f" Resuming from checkpoint...")
results = checkpoint_data['results']
completed_ids = set(checkpoint_data['completed_ids'])
start_time = checkpoint_data['start_time']
resume_from = len(completed_ids)
else:
completed_ids = set()
results = {'correct': [], 'incorrect': [], 'timeouts': [], 'errors': []}
start_time = time.time()
resume_from = 0
else:
completed_ids = set()
results = {'correct': [], 'incorrect': [], 'timeouts': [], 'errors': []}
start_time = time.time()
resume_from = 0
# Load dataset
dataset_path = Path(config_path).parent / bench_cfg['dataset']
if not dataset_path.exists():
print(f"Dataset not found: {dataset_path}", file=sys.stderr)
sys.exit(1)
df = pd.read_parquet(dataset_path)
questions = df.to_dict('records')
total = len(questions)
max_q = bench_cfg.get('max_questions', 0) or total
if max_q < total:
import random
random.seed(bench_cfg.get('seed', 42))
random.shuffle(questions)
questions = questions[:max_q]
# Filter out already completed questions when resuming
if resume_from > 0:
remaining = [q for q in questions if q.get('id', '') not in completed_ids]
skipped = len(questions) - len(remaining)
questions = remaining
print(f" Skipped {skipped} completed questions. {len(questions)} remaining.\n")
print(f"\n{'='*60}")
print(f" MCQ Benchmark")
print(f"{'='*60}")
print(f" Model: {model_cfg['name']}")
print(f" Endpoint: {model_cfg['endpoint']}")
print(f" Timeout: {model_cfg['timeout']}s")
print(f" Max tokens: {model_cfg['max_tokens']}")
print(f" Temperature: {model_cfg.get('temperature', 0.0)}")
print(f" Questions: {len(questions) + resume_from} / {total}")
if resume_from > 0:
print(f" Resumed: {resume_from} completed")
print(f"{'='*60}\n")
# Determine checkpoint file path
if not checkpoint_path:
timestamp = datetime.now().strftime('%Y%m%d_%H%M%S')
checkpoint_path = output_dir / f"{safe_name}_{timestamp}.checkpoint.json"
global_idx = resume_from
for i, q in enumerate(questions, 1):
qid = q.get('id', f'unknown_{i}')
question_text = q.get('text', '?')
golden_key = q.get('golden_key', '?')
options = q.get('options', {})
global_idx += 1
display_q = question_text[:80] + ('...' if len(question_text) > 80 else '')
prompt = build_prompt(q)
response = None
timed_out = False
for attempt in range(1 + bench_cfg.get('max_retries', 0)):
response = query_model(prompt, config)
if response is None:
timed_out = True
continue
break
is_correct = False
model_choice = ''
if timed_out:
results['timeouts'].append({
'id': qid,
'question': question_text,
'expected_key': golden_key,
'options': options,
})
elif answers_match(golden_key, response or '', options):
is_correct = True
results['correct'].append({
'id': qid,
'question': question_text,
'expected_key': golden_key,
'model_response': (response or '')[:200],
})
else:
model_choice = extract_choice(response or '')
results['incorrect'].append({
'id': qid,
'question': question_text,
'expected_key': golden_key,
'model_response': (response or '')[:500],
'model_choice': model_choice,
'options': options,
})
# Running accuracy
answered = len(results['correct']) + len(results['incorrect']) + len(results['timeouts'])
acc = (len(results['correct']) / answered * 100) if answered > 0 else 0.0
if timed_out:
label = 'TIMEOUT'
elif is_correct:
label = 'CORRECT'
else:
label = f"WRONG (expected: {golden_key}, got: {model_choice or '?'})"
status = f" [{global_idx}/{len(questions) + resume_from}] {display_q} {label} [{acc:.1f}%]"
print(f"\r{status}")
# Save checkpoint after each question
completed_ids.add(qid)
save_checkpoint(checkpoint_path, results, list(completed_ids), start_time)
if bench_cfg.get('delay_between_requests', 0) > 0:
time.sleep(bench_cfg['delay_between_requests'])
# Build and save final report
report = build_report(config, results, start_time)
timestamp = datetime.now().strftime('%Y%m%d_%H%M%S')
filename = f"{safe_name}_{timestamp}.json"
output_path = output_dir / filename
with open(output_path, 'w') as f:
json.dump(report, f, indent=2, ensure_ascii=False)
# Clean up checkpoint
if checkpoint_path.exists():
checkpoint_path.unlink()
print_summary(report, output_path)
return report
def main():
parser = argparse.ArgumentParser(description='MCQ LLM Benchmark')
parser.add_argument('--config', '-c', default='config.yaml',
help='Path to config.yaml (default: config.yaml)')
parser.add_argument('--resume', '-r', action='store_true',
help='Resume from the latest checkpoint')
parser.add_argument('--fresh', '-f', action='store_true',
help='Start a new run, ignoring existing checkpoint')
args = parser.parse_args()
config_path = Path(args.config)
if not config_path.exists():
print(f"Config not found: {config_path}")
print(f" Copy config.example.yaml to config.yaml and fill in your values.", file=sys.stderr)
sys.exit(1)
run_benchmark(str(config_path), resume=args.resume, fresh=args.fresh)
if __name__ == '__main__':
main()
config.example.yaml
+31
View File
@@ -0,0 +1,31 @@
# MCQ Benchmark - Configuration
# Copy to config.yaml and fill in your values.
# Model configuration
model:
name: "your-model-name"
endpoint: "http://localhost:8080/v1"
api_key: "not-needed"
timeout: 300
max_tokens: 4096
temperature: 0.0
# Benchmark settings
benchmark:
# Path to the parquet dataset file
dataset: "train-00000-of-00001.parquet"
# Number of questions to run (0 = all)
max_questions: 0
# Random seed for reproducibility when max_questions < total
seed: 42
# How many times to retry a failed request
max_retries: 1
# Delay in seconds between requests (0 = none)
delay_between_requests: 0
# Output directory for results
output_dir: "runs"