LightRAG/tests/gpt5_nano_compatibility/test_gpt5_nano_compatibility.py
2025-12-05 14:31:13 +08:00

313 lines
10 KiB
Python

#!/usr/bin/env python3
"""
Test script to verify gpt-5-nano compatibility with LightRAG.
This script validates that:
1. gpt-5-nano parameter handling works correctly (max_completion_tokens conversion)
2. Temperature parameter is properly handled for gpt-5-nano
3. Embeddings work with gpt-5-nano configuration
4. Entity extraction works with gpt-5-nano
5. Full pipeline works end-to-end
Requires:
- OPENAI_API_KEY environment variable set
- LLM_MODEL set to gpt-5-nano or specified via argument
Run standalone: python test_gpt5_nano_compatibility.py
Run via pytest: pytest test_gpt5_nano_compatibility.py -v (some tests skipped without API key)
"""
import os
import sys
import asyncio
import logging
import pytest
# Setup logging
logging.basicConfig(
level=logging.INFO, format="%(asctime)s - %(name)s - %(levelname)s - %(message)s"
)
logger = logging.getLogger(__name__)
# Add the repo to path
sys.path.insert(
0, os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
)
from lightrag.llm.openai import (
openai_complete_if_cache,
openai_embed,
_normalize_openai_kwargs_for_model,
)
# Define markers for different test types
NEEDS_API_KEY = pytest.mark.skipif(
not os.getenv("OPENAI_API_KEY") and not os.getenv("LLM_BINDING_API_KEY"),
reason="OPENAI_API_KEY or LLM_BINDING_API_KEY not set",
)
@pytest.mark.asyncio
async def test_parameter_normalization():
"""Test 1: Parameter normalization for gpt-5-nano"""
logger.info("=" * 60)
logger.info("TEST 1: Parameter normalization for gpt-5-nano")
logger.info("=" * 60)
try:
# Test case 1a: max_tokens conversion to max_completion_tokens with buffer
kwargs = {"max_tokens": 500, "temperature": 0.7, "top_p": 0.9}
original_kwargs = kwargs.copy()
_normalize_openai_kwargs_for_model("gpt-5-nano", kwargs)
logger.info(f"Input kwargs: {original_kwargs}")
logger.info(f"Output kwargs: {kwargs}")
assert (
"max_completion_tokens" in kwargs
), "max_tokens should be converted to max_completion_tokens"
assert (
kwargs["max_completion_tokens"] >= 500
), "max_completion_tokens should be at least original value (500)"
assert "max_tokens" not in kwargs, "max_tokens should be removed"
assert (
"temperature" not in kwargs
), "temperature should be removed for gpt-5-nano"
assert "top_p" in kwargs, "top_p should be preserved"
logger.info(
f"✅ Test 1a passed: max_tokens → max_completion_tokens conversion works (buffered from 500 to {kwargs['max_completion_tokens']})"
)
# Test case 1b: Both max_tokens and max_completion_tokens (edge case)
kwargs = {"max_tokens": 200, "max_completion_tokens": 300, "temperature": 0.5}
original_kwargs = kwargs.copy()
_normalize_openai_kwargs_for_model("gpt-5-nano", kwargs)
logger.info(f"Input kwargs (both max params): {original_kwargs}")
logger.info(f"Output kwargs: {kwargs}")
assert "max_tokens" not in kwargs, "max_tokens should be removed"
assert "max_completion_tokens" in kwargs, "max_completion_tokens should be kept"
assert "temperature" not in kwargs, "temperature should be removed"
logger.info("✅ Test 1b passed: Both max parameters handled correctly")
# Test case 1c: Non-gpt5 models shouldn't change
kwargs = {"max_tokens": 500, "temperature": 0.7}
original_kwargs = kwargs.copy()
_normalize_openai_kwargs_for_model("gpt-4o-mini", kwargs)
logger.info(f"Input kwargs (gpt-4o-mini): {original_kwargs}")
logger.info(f"Output kwargs: {kwargs}")
assert "max_tokens" in kwargs, "max_tokens should be preserved for gpt-4o-mini"
assert kwargs["max_tokens"] == 500, "max_tokens value should be unchanged"
assert (
"temperature" in kwargs
), "temperature should be preserved for gpt-4o-mini"
logger.info("✅ Test 1c passed: Non-gpt5 models are unchanged")
logger.info("✅ TEST 1 PASSED: Parameter normalization works correctly\n")
return True
except Exception as e:
logger.error(f"❌ TEST 1 FAILED: {e}")
import traceback
traceback.print_exc()
return False
@pytest.mark.asyncio
@NEEDS_API_KEY
async def test_embeddings():
"""Test 2: Embeddings generation"""
logger.info("=" * 60)
logger.info("TEST 2: Embeddings generation")
logger.info("=" * 60)
try:
texts = ["Hello world", "This is a test"]
model = os.getenv("EMBEDDING_MODEL", "text-embedding-3-small")
logger.info(f"Generating embeddings with model: {model}")
embeddings = await openai_embed(texts, model=model)
logger.info(f"Generated {len(embeddings)} embeddings")
logger.info(f"First embedding shape: {len(embeddings[0])}")
assert len(embeddings) == len(texts), "Should get one embedding per text"
assert len(embeddings[0]) > 0, "Embeddings should not be empty"
logger.info("✅ TEST 2 PASSED: Embeddings generation works\n")
return True
except Exception as e:
logger.error(f"❌ TEST 2 FAILED: {e}")
import traceback
traceback.print_exc()
return False
@pytest.mark.asyncio
@pytest.mark.skipif(not os.getenv("OPENAI_API_KEY"), reason="OPENAI_API_KEY not set")
async def test_simple_completion():
"""Test 3: Simple LLM completion with gpt-5-nano"""
logger.info("=" * 60)
logger.info("TEST 3: Simple LLM completion with gpt-5-nano")
logger.info("=" * 60)
try:
model = os.getenv("LLM_MODEL", "gpt-5-nano")
logger.info(f"Testing completion with model: {model}")
# Enable verbose debug logging
import logging as py_logging
py_logging.getLogger("openai").setLevel(py_logging.DEBUG)
# Test without custom temperature (gpt-5-nano requirement)
response = await openai_complete_if_cache(
model=model,
prompt="Say hello in one word",
system_prompt="You are a helpful assistant.",
max_completion_tokens=20,
)
logger.info(f"Response: {response}")
assert len(response) > 0, "Response should not be empty"
logger.info("✅ TEST 3 PASSED: Simple completion works\n")
return True
except Exception as e:
logger.error(f"❌ TEST 3 FAILED: {e}")
import traceback
traceback.print_exc()
return False
@pytest.mark.asyncio
@pytest.mark.skipif(not os.getenv("OPENAI_API_KEY"), reason="OPENAI_API_KEY not set")
async def test_extraction_with_gpt5nano():
"""Test 4: Entity extraction style task"""
logger.info("=" * 60)
logger.info("TEST 4: Entity extraction style task")
logger.info("=" * 60)
try:
model = os.getenv("LLM_MODEL", "gpt-5-nano")
prompt = """Extract the entities from this text:
Apple Inc. was founded by Steve Jobs in 1976.
Return as JSON with keys: company, person, year."""
logger.info(f"Testing extraction with model: {model}")
response = await openai_complete_if_cache(
model=model,
prompt=prompt,
system_prompt="You are an entity extraction assistant. Always respond in valid JSON.",
max_completion_tokens=100,
)
logger.info(f"Response: {response}")
assert len(response) > 0, "Response should not be empty"
assert "Apple" in response or "apple" in response, "Should mention Apple"
logger.info("✅ TEST 4 PASSED: Entity extraction works\n")
return True
except Exception as e:
logger.error(f"❌ TEST 4 FAILED: {e}")
import traceback
traceback.print_exc()
return False
@pytest.mark.asyncio
async def test_config_loading():
"""Test 5: Configuration loading from .env"""
logger.info("=" * 60)
logger.info("TEST 5: Configuration loading from .env")
logger.info("=" * 60)
llm_model = os.getenv("LLM_MODEL", "not-set")
llm_binding = os.getenv("LLM_BINDING", "not-set")
embedding_model = os.getenv("EMBEDDING_MODEL", "not-set")
embedding_binding = os.getenv("EMBEDDING_BINDING", "not-set")
logger.info(f"LLM_MODEL: {llm_model}")
logger.info(f"LLM_BINDING: {llm_binding}")
logger.info(f"EMBEDDING_MODEL: {embedding_model}")
logger.info(f"EMBEDDING_BINDING: {embedding_binding}")
# Verify we're using OpenAI
assert (
embedding_binding == "openai" or embedding_binding == "not-set"
), "EMBEDDING_BINDING should be openai"
assert (
llm_binding == "openai" or llm_binding == "not-set"
), "LLM_BINDING should be openai"
logger.info("✅ TEST 5 PASSED: Configuration loaded correctly\n")
return True
async def _run_all_tests():
"""Run all tests (internal helper, not picked up by pytest)"""
logger.info("\n" + "=" * 60)
logger.info("GPT-5-NANO COMPATIBILITY TEST SUITE")
logger.info("=" * 60 + "\n")
# Check prerequisites
api_key = os.getenv("OPENAI_API_KEY")
if not api_key:
logger.error("❌ OPENAI_API_KEY environment variable not set")
return False
results = {
"Parameter Normalization": await test_parameter_normalization(),
"Configuration Loading": await test_config_loading(),
"Embeddings": await test_embeddings(),
"Simple Completion": await test_simple_completion(),
"Entity Extraction": await test_extraction_with_gpt5nano(),
}
# Summary
logger.info("=" * 60)
logger.info("TEST SUMMARY")
logger.info("=" * 60)
for test_name, result in results.items():
status = "✅ PASSED" if result else "❌ FAILED"
logger.info(f"{test_name}: {status}")
all_passed = all(results.values())
if all_passed:
logger.info("\n" + "=" * 60)
logger.info("🎉 ALL TESTS PASSED")
logger.info("=" * 60)
else:
logger.info("\n" + "=" * 60)
logger.info("⚠️ SOME TESTS FAILED")
logger.info("=" * 60)
return all_passed
if __name__ == "__main__":
# Load environment from .env file
from dotenv import load_dotenv
load_dotenv(dotenv_path=".env", override=False)
# Run tests
success = asyncio.run(_run_all_tests())
sys.exit(0 if success else 1)