dataset-builder / data3 /generate_problems_batch.py
SunDou's picture
Upload data3/generate_problems_batch.py with huggingface_hub
0003466 verified
#!/usr/bin/env python3
"""
Generate programming problems from function_dataset_v2.csv using OpenAI Batch API.
Batch API offers 50% cost savings compared to standard API.
"""
import csv
import json
import os
import sys
from openai import OpenAI
from datetime import datetime
from typing import Dict, Optional, List
import time
# Configuration
MODEL_NAME = "gpt-4o-mini"
MIN_RELEVANCE_SCORE = 60
MAX_BUDGET_USD = 10.0
# OpenAI Batch API pricing (50% off standard pricing)
# Official pricing: https://openai.com/api/pricing/
BATCH_PRICING = {
# GPT-5 series with Batch API discount
"gpt-5.2": {
"input": 0.875 / 1_000_000, # $0.875 per 1M (50% off $1.75)
"output": 7.00 / 1_000_000, # $7.00 per 1M (50% off $14.00)
},
"gpt-5.1": {
"input": 0.625 / 1_000_000, # $0.625 per 1M (50% off $1.25)
"output": 5.00 / 1_000_000, # $5.00 per 1M (50% off $10.00)
},
"gpt-5": {
"input": 0.625 / 1_000_000, # $0.625 per 1M (50% off $1.25)
"output": 5.00 / 1_000_000, # $5.00 per 1M (50% off $10.00)
},
"gpt-5-mini": {
"input": 0.125 / 1_000_000, # $0.125 per 1M (50% off $0.25)
"output": 1.00 / 1_000_000, # $1.00 per 1M (50% off $2.00)
},
"gpt-5-nano": {
"input": 0.025 / 1_000_000, # $0.025 per 1M (50% off $0.05)
"output": 0.20 / 1_000_000, # $0.20 per 1M (50% off $0.40)
},
# GPT-4o series with Batch API discount
"gpt-4o": {
"input": 1.25 / 1_000_000, # $1.25 per 1M (50% off $2.50)
"output": 5.00 / 1_000_000, # $5.00 per 1M (50% off $10.00)
},
"gpt-4o-2024-05-13": {
"input": 2.50 / 1_000_000, # $2.50 per 1M (50% off $5.00)
"output": 7.50 / 1_000_000, # $7.50 per 1M (50% off $15.00)
},
"gpt-4o-mini": {
"input": 0.075 / 1_000_000, # $0.075 per 1M (50% off $0.15)
"output": 0.30 / 1_000_000, # $0.30 per 1M (50% off $0.60)
},
# GPT-4 Turbo
"gpt-4-turbo": {
"input": 5.00 / 1_000_000, # $5.00 per 1M (50% off $10.00)
"output": 15.00 / 1_000_000, # $15.00 per 1M (50% off $30.00)
},
# GPT-3.5 Turbo
"gpt-3.5-turbo": {
"input": 0.25 / 1_000_000, # $0.25 per 1M (50% off $0.50)
"output": 0.75 / 1_000_000, # $0.75 per 1M (50% off $1.50)
},
}
PROMPT_TEMPLATE = """You are an expert in scientific computing and computational chemistry/biology/physics. Please create a high-quality programming problem inspired by the following code snippet from a real scientific computing project.
The problem should focus on scientific computing concepts such as:
- Numerical algorithms and simulations
- Data analysis and visualization
- Mathematical modeling
- Scientific data processing
- Computational methods in chemistry, biology, or physics
Code snippet for inspiration:
```python
{code}
```
Present your output in two distinct sections:
[Problem Description]
Create a **completely self-contained** problem description that:
- Does NOT directly reference the code snippet above
- Provides all necessary context and background
- Clearly states what needs to be implemented
- Specifies input/output format and constraints
- Is inspired by the scientific computing concepts in the code but creates a NEW, interesting problem
- Assumes common programming knowledge but explains any domain-specific concepts
[Solution]
Provide a comprehensive, **correct** Python solution that:
- Accurately solves the problem described
- Includes clear comments explaining the approach
- Uses appropriate scientific computing libraries (numpy, scipy, etc.) when relevant
- Is complete and runnable
- Follows best practices for scientific computing
Remember: The problem should be INSPIRED by the code, not a direct copy. Create something educational and interesting for scientific computing practitioners."""
class BatchAPIClient:
"""Client for OpenAI Batch API with cost tracking."""
def __init__(self, model_name: str = MODEL_NAME, api_key: Optional[str] = None):
"""Initialize OpenAI Batch API client.
Args:
model_name: Name of the OpenAI model to use
api_key: OpenAI API key (if None, will use OPENAI_API_KEY env variable)
"""
self.model_name = model_name
self.client = OpenAI(api_key=api_key)
# Get pricing for the model (Batch API is 50% off)
if model_name in BATCH_PRICING:
self.input_price = BATCH_PRICING[model_name]["input"]
self.output_price = BATCH_PRICING[model_name]["output"]
else:
print(f"Warning: No Batch pricing info for {model_name}, using gpt-4o-mini prices")
self.input_price = BATCH_PRICING["gpt-4o-mini"]["input"]
self.output_price = BATCH_PRICING["gpt-4o-mini"]["output"]
print(f"๐Ÿ“Š Batch API Pricing (50% off standard rates):")
print(f" Input: ${self.input_price * 1_000_000:.4f} per 1M tokens")
print(f" Output: ${self.output_price * 1_000_000:.4f} per 1M tokens")
print()
def create_batch_file(self, requests: List[Dict], output_path: str) -> str:
"""Create a JSONL file for batch processing.
Args:
requests: List of request dictionaries
output_path: Path to save the JSONL file
Returns:
Path to the created file
"""
with open(output_path, 'w', encoding='utf-8') as f:
for req in requests:
f.write(json.dumps(req, ensure_ascii=False) + '\n')
print(f"โœ… Created batch file: {output_path}")
print(f" Total requests: {len(requests)}")
return output_path
def upload_batch_file(self, file_path: str) -> str:
"""Upload batch file to OpenAI.
Args:
file_path: Path to the JSONL file
Returns:
File ID
"""
print(f"โฌ†๏ธ Uploading batch file to OpenAI...")
with open(file_path, 'rb') as f:
batch_file = self.client.files.create(
file=f,
purpose='batch'
)
print(f"โœ… File uploaded: {batch_file.id}")
return batch_file.id
def create_batch(self, file_id: str, description: Optional[str] = None) -> str:
"""Create a batch job.
Args:
file_id: ID of the uploaded file
description: Optional description for the batch
Returns:
Batch ID
"""
print(f"๐Ÿš€ Creating batch job...")
batch = self.client.batches.create(
input_file_id=file_id,
endpoint="/v1/chat/completions",
completion_window="24h",
metadata={
"description": description or "Programming problems generation",
"created_at": datetime.now().isoformat()
}
)
print(f"โœ… Batch created: {batch.id}")
print(f" Status: {batch.status}")
print(f" Total requests: {batch.request_counts.total}")
return batch.id
def check_batch_status(self, batch_id: str) -> Dict:
"""Check the status of a batch job.
Args:
batch_id: ID of the batch
Returns:
Batch status information
"""
batch = self.client.batches.retrieve(batch_id)
status_info = {
'id': batch.id,
'status': batch.status,
'created_at': batch.created_at,
'completed_at': batch.completed_at,
'failed_at': batch.failed_at,
'expired_at': batch.expired_at,
'request_counts': {
'total': batch.request_counts.total,
'completed': batch.request_counts.completed,
'failed': batch.request_counts.failed,
},
'output_file_id': batch.output_file_id,
'error_file_id': batch.error_file_id,
}
return status_info
def download_results(self, file_id: str, output_path: str):
"""Download batch results.
Args:
file_id: ID of the output file
output_path: Path to save the results
"""
print(f"โฌ‡๏ธ Downloading results...")
content = self.client.files.content(file_id)
with open(output_path, 'wb') as f:
f.write(content.content)
print(f"โœ… Results saved to: {output_path}")
def estimate_cost(self, num_requests: int, avg_input_tokens: int, avg_output_tokens: int) -> Dict:
"""Estimate the cost of a batch job.
Args:
num_requests: Number of requests
avg_input_tokens: Average input tokens per request
avg_output_tokens: Average output tokens per request
Returns:
Cost estimation dictionary
"""
total_input_tokens = num_requests * avg_input_tokens
total_output_tokens = num_requests * avg_output_tokens
input_cost = total_input_tokens * self.input_price
output_cost = total_output_tokens * self.output_price
total_cost = input_cost + output_cost
# Compare with standard API (2x the batch price)
standard_cost = total_cost * 2
savings = standard_cost - total_cost
return {
'num_requests': num_requests,
'total_input_tokens': total_input_tokens,
'total_output_tokens': total_output_tokens,
'total_tokens': total_input_tokens + total_output_tokens,
'input_cost': input_cost,
'output_cost': output_cost,
'total_cost': total_cost,
'standard_api_cost': standard_cost,
'savings': savings,
'savings_percentage': 50.0
}
def prepare_batch_requests(
input_file: str,
min_score: int = MIN_RELEVANCE_SCORE,
max_samples: Optional[int] = None,
start_from: int = 0,
) -> List[Dict]:
"""Prepare batch requests from function dataset.
Args:
input_file: Path to function_dataset_v2.csv
min_score: Minimum relevance score to process
max_samples: Maximum number of samples to process
start_from: Skip first N rows
Returns:
List of batch request dictionaries
"""
print(f"๐Ÿ“‹ Preparing batch requests...")
print(f" Input: {input_file}")
print(f" Min Score: {min_score}")
if max_samples:
print(f" Max Samples: {max_samples}")
print()
requests = []
total_rows = 0
skipped_low_score = 0
skipped_no_code = 0
with open(input_file, 'r', encoding='utf-8') as infile:
reader = csv.DictReader(infile)
for row in reader:
total_rows += 1
# Skip if resuming
if total_rows <= start_from:
continue
# Check if we've reached max samples
if max_samples and len(requests) >= max_samples:
break
# Filter by relevance score
try:
relevance_score = int(row.get('relevance_score', 0))
except (ValueError, TypeError):
relevance_score = 0
if relevance_score < min_score:
skipped_low_score += 1
continue
# Get function content
function_content = row.get('function_content', '').strip()
if not function_content or len(function_content) < 50:
skipped_no_code += 1
continue
# Prepare metadata (OpenAI Batch API requires all metadata values to be strings)
metadata = {
'original_index': str(row.get('original_index', '')),
'function_name': str(row.get('function_name', '')),
'repo_name': str(row.get('repo_name', '')),
'path': str(row.get('path', '')),
'language': str(row.get('language', '')),
'relevance_score': str(relevance_score), # Convert to string!
'function_start_line': str(row.get('function_start_line', '')),
'function_end_line': str(row.get('function_end_line', '')),
}
# Generate prompt
prompt = PROMPT_TEMPLATE.format(code=function_content)
# Create batch request in OpenAI Batch API format
request = {
"custom_id": f"request-{len(requests)}",
"method": "POST",
"url": "/v1/chat/completions",
"body": {
"model": MODEL_NAME,
"messages": [
{
"role": "system",
"content": "You are an expert in scientific computing and programming education."
},
{
"role": "user",
"content": prompt
}
],
"temperature": 0.7,
"metadata": metadata # All values are now strings
}
}
requests.append(request)
print(f"โœ… Prepared {len(requests)} requests")
print(f" Total rows: {total_rows}")
print(f" Skipped (low score): {skipped_low_score}")
print(f" Skipped (no/short code): {skipped_no_code}")
print()
return requests
def process_batch_results(
results_file: str,
output_file: str,
model_name: str,
input_price: float,
output_price: float,
requests_file: Optional[str] = None
):
"""Process batch results and save to JSONL format.
Args:
results_file: Path to batch results file
output_file: Path to output JSONL file
model_name: Model name used
input_price: Input token price
output_price: Output token price
requests_file: Optional path to original batch requests file (to restore prompts)
"""
print(f"๐Ÿ“Š Processing batch results...")
# Load prompts from requests file if provided
prompts_map = {}
if requests_file and os.path.exists(requests_file):
print(f" Loading prompts from: {requests_file}")
with open(requests_file, 'r', encoding='utf-8') as f:
for line in f:
req = json.loads(line)
custom_id = req['custom_id']
# Extract prompt from messages
for msg in req['body']['messages']:
if msg['role'] == 'user':
prompts_map[custom_id] = msg['content']
break
print(f" Loaded {len(prompts_map)} prompts")
processed = 0
errors = 0
total_input_tokens = 0
total_output_tokens = 0
total_cost = 0.0
with open(results_file, 'r', encoding='utf-8') as infile, \
open(output_file, 'w', encoding='utf-8') as outfile:
for line in infile:
batch_result = json.loads(line)
# Check if request was successful
if batch_result.get('error'):
errors += 1
print(f"โŒ Error in {batch_result['custom_id']}: {batch_result['error']}")
continue
response = batch_result['response']
custom_id = batch_result['custom_id']
# Extract usage information
usage = response['body']['usage']
input_tokens = usage['prompt_tokens']
output_tokens = usage['completion_tokens']
# Calculate cost
input_cost = input_tokens * input_price
output_cost = output_tokens * output_price
request_cost = input_cost + output_cost
# Update totals
total_input_tokens += input_tokens
total_output_tokens += output_tokens
total_cost += request_cost
# Get metadata from the original request
metadata = response['body'].get('metadata', {})
# Extract the response text
response_text = response['body']['choices'][0]['message']['content']
# Build result - include prompt if available
result = {
'metadata': metadata,
'response': response_text,
'usage': {
'input_tokens': input_tokens,
'output_tokens': output_tokens,
'total_tokens': input_tokens + output_tokens,
'input_cost': input_cost,
'output_cost': output_cost,
'request_cost': request_cost
},
'model': model_name,
'timestamp': datetime.now().isoformat(),
'custom_id': custom_id
}
# Add prompt if we have it
if custom_id in prompts_map:
result['prompt'] = prompts_map[custom_id]
outfile.write(json.dumps(result, ensure_ascii=False) + '\n')
processed += 1
print(f"\nโœ… Processed {processed} results")
print(f" Errors: {errors}")
print()
# Print usage summary
print("=" * 70)
print("BATCH API USAGE SUMMARY")
print("=" * 70)
print(f"Model: {model_name}")
print(f"Total Requests: {processed}")
print(f"Total Input Tokens: {total_input_tokens:,}")
print(f"Total Output Tokens: {total_output_tokens:,}")
print(f"Total Tokens: {total_input_tokens + total_output_tokens:,}")
print(f"\nBatch API Cost: ${total_cost:.6f}")
print(f"Standard API Cost: ${total_cost * 2:.6f}")
print(f"Savings (50%): ${total_cost:.6f}")
print("=" * 70)
def main():
import argparse
parser = argparse.ArgumentParser(
description='Generate programming problems using OpenAI Batch API (50% cost savings)'
)
subparsers = parser.add_subparsers(dest='command', help='Command to run')
# Prepare command
prepare_parser = subparsers.add_parser('prepare', help='Prepare batch requests')
prepare_parser.add_argument('--input', default='function_dataset_v2.csv')
prepare_parser.add_argument('--output', default='batch_requests.jsonl')
prepare_parser.add_argument('--min-score', type=int, default=MIN_RELEVANCE_SCORE)
prepare_parser.add_argument('--max-samples', type=int, default=None)
prepare_parser.add_argument('--start-from', type=int, default=0)
prepare_parser.add_argument('--model', default=MODEL_NAME)
# Submit command
submit_parser = subparsers.add_parser('submit', help='Submit batch job to OpenAI')
submit_parser.add_argument('--input', default='batch_requests.jsonl')
submit_parser.add_argument('--model', default=MODEL_NAME)
submit_parser.add_argument('--description', default='Programming problems generation')
# Status command
status_parser = subparsers.add_parser('status', help='Check batch job status')
status_parser.add_argument('batch_id', help='Batch ID to check')
# Download command
download_parser = subparsers.add_parser('download', help='Download batch results')
download_parser.add_argument('batch_id', help='Batch ID to download')
download_parser.add_argument('--output', default='batch_results.jsonl')
# Process command
process_parser = subparsers.add_parser('process', help='Process downloaded results')
process_parser.add_argument('--input', default='batch_results.jsonl')
process_parser.add_argument('--output', default='programming_problems_batch.jsonl')
process_parser.add_argument('--model', default=MODEL_NAME)
process_parser.add_argument('--requests', default='batch_requests_full.jsonl',
help='Original batch requests file (to restore prompts)')
# Estimate command
estimate_parser = subparsers.add_parser('estimate', help='Estimate batch cost')
estimate_parser.add_argument('--num-requests', type=int, required=True)
estimate_parser.add_argument('--avg-input-tokens', type=int, default=1917)
estimate_parser.add_argument('--avg-output-tokens', type=int, default=2552)
estimate_parser.add_argument('--model', default=MODEL_NAME)
args = parser.parse_args()
if not args.command:
parser.print_help()
sys.exit(1)
# Check API key
if not os.getenv('OPENAI_API_KEY'):
print("โŒ Error: OPENAI_API_KEY environment variable not set.")
print(" Please set it with: export OPENAI_API_KEY='your-api-key'")
sys.exit(1)
client = BatchAPIClient(model_name=args.model if hasattr(args, 'model') else MODEL_NAME)
if args.command == 'prepare':
requests = prepare_batch_requests(
input_file=args.input,
min_score=args.min_score,
max_samples=args.max_samples,
start_from=args.start_from
)
client.create_batch_file(requests, args.output)
# Estimate cost
print("\n๐Ÿ’ฐ Cost Estimation:")
estimate = client.estimate_cost(
num_requests=len(requests),
avg_input_tokens=1917, # From your test
avg_output_tokens=2552 # From your test
)
print(f" Estimated Batch API Cost: ${estimate['total_cost']:.2f}")
print(f" Standard API Cost: ${estimate['standard_api_cost']:.2f}")
print(f" Savings (50%): ${estimate['savings']:.2f}")
print()
elif args.command == 'submit':
file_id = client.upload_batch_file(args.input)
batch_id = client.create_batch(file_id, args.description)
print(f"\n๐Ÿ“ Save this Batch ID: {batch_id}")
print(f" Check status with: python3 {sys.argv[0]} status {batch_id}")
elif args.command == 'status':
status = client.check_batch_status(args.batch_id)
print("\n๐Ÿ“Š Batch Status:")
print(f" ID: {status['id']}")
print(f" Status: {status['status']}")
print(f" Total: {status['request_counts']['total']}")
print(f" Completed: {status['request_counts']['completed']}")
print(f" Failed: {status['request_counts']['failed']}")
if status['status'] == 'completed':
print(f"\nโœ… Batch completed!")
print(f" Download with: python3 {sys.argv[0]} download {args.batch_id}")
elif status['status'] == 'failed':
print(f"\nโŒ Batch failed!")
else:
print(f"\nโณ Batch is still processing...")
elif args.command == 'download':
status = client.check_batch_status(args.batch_id)
if status['status'] != 'completed':
print(f"โŒ Batch is not completed yet (status: {status['status']})")
sys.exit(1)
client.download_results(status['output_file_id'], args.output)
print(f"\nโœ… Downloaded to: {args.output}")
print(f" Process with: python3 {sys.argv[0]} process --input {args.output}")
elif args.command == 'process':
process_batch_results(
results_file=args.input,
output_file=args.output,
model_name=args.model,
input_price=client.input_price,
output_price=client.output_price,
requests_file=args.requests
)
print(f"\nโœ… Final results saved to: {args.output}")
elif args.command == 'estimate':
estimate = client.estimate_cost(
num_requests=args.num_requests,
avg_input_tokens=args.avg_input_tokens,
avg_output_tokens=args.avg_output_tokens
)
print("\n๐Ÿ’ฐ COST ESTIMATION")
print("=" * 70)
print(f"Number of Requests: {estimate['num_requests']:,}")
print(f"Total Input Tokens: {estimate['total_input_tokens']:,}")
print(f"Total Output Tokens: {estimate['total_output_tokens']:,}")
print(f"Total Tokens: {estimate['total_tokens']:,}")
print()
print(f"Batch API Cost: ${estimate['total_cost']:.2f}")
print(f"Standard API Cost: ${estimate['standard_api_cost']:.2f}")
print(f"๐Ÿ’ฐ Savings (50%): ${estimate['savings']:.2f}")
print("=" * 70)
if __name__ == "__main__":
main()