LLM Performance Benchmark

Benchmark the performance of OpenAI compatible APIs in terms of Time to First Token (commonly referred to as latency) and Output Tokens per Second.

Background

Benchmarking script used by Artificial Analysis for our performance benchmarks.

For performance benchmarks across >150 endpoints, including across varied prompt lengths & parallel queries, visit: Artificial Analysis API Provider Peformance Leaderboard

Install

# Install from PyPI
pip install llm_performance_benchmark

Usage

from llm_performance_benchmark import llm_performance_benchmark

# Configure endpoint
import os
from dotenv import load_dotenv
load_dotenv()
api_key = os.getenv("OPENAI_API_KEY")
base_url = "https://api.openai.com/v1"
model_id = "gpt-4-turbo"

# Configure prompt
user_prompt = "What is the purpose of life?"
system_prompt = "You are a helpful assistant."

# Run benchmark
result = llm_performance_benchmark(model_id=model_id, 
                                   user_prompt=user_prompt, 
                                   system_prompt=system_prompt, 
                                   api_key=api_key, 
                                   base_url=base_url,
                                   print_response=True)
print(result)
""" Example response:
{
    'total_time': 8.9, 
    'time_to_first_token': 0.6, 
    'output_tokens_per_second': 39.4, 
    'tokens_per_second_across_total_request': 36.4,
    'response_text: "..."
}
"""

Definitions of output metrics

For definitions of what the metrics represent, see the following https://artificialanalysis.ai/methodology page. To standardize token counting across different models, text is tokenized as cl100k_base tokens (used by GPT-4) with Tiktoken.

License

MIT License

Copyright (c) 2024 Artificial Analysis, Inc.

See LICENSE file for further details