fix(openai): resolve vLLM compatibility issue with ChatOpenAI (#32252)

[20ns]

Overview

This PR fixes the vLLM compatibility issue reported in langchain-ai#32252 where users encountered a TypeError: "Received response with null value for 'choices'" when using vLLM with LangChain-OpenAI integration.

Problem Description

The issue occurred because some OpenAI-compatible APIs like vLLM return valid response objects, but the OpenAI client library's model_dump() method sometimes fails to properly serialize the choices field, returning None instead of the actual choices array. This caused the _create_chat_result method in BaseChatOpenAI to raise a TypeError.

Solution

Added a fallback mechanism in the _create_chat_result method that:

1. Primary Path: First attempts to get choices from response.model_dump() as before
2. Fallback Path: If model_dump() returns choices: None, attempts to access choices directly from the response object via response.choices
3. Graceful Conversion: Converts the raw choice objects to the expected dictionary format using model_dump() or __dict__ fallback
4. Backward Compatibility: Maintains the original error behavior when choices are truly unavailable

Changes Made

Core Fix

• File: libs/partners/openai/langchain_openai/chat_models/base.py
• Method: _create_chat_result() (lines ~1208-1230)
• Change: Added fallback logic to handle cases where model_dump() returns choices: None

Test Coverage

• File: libs/partners/openai/tests/unit_tests/chat_models/test_vllm_compatibility.py (new)
• Coverage: 6 comprehensive test cases covering:
  • ✅ Main Fix Test: vLLM response where model_dump() fails but response.choices works
  • ✅ Backward Compatibility: Cases where choices are truly unavailable still raise errors
  • ✅ Working Cases: Normal responses that work through model_dump()
  • ✅ Dict Responses: Direct dictionary responses (bypass issue entirely)
  • ✅ Edge Cases: Null choices and missing choices key

Testing

# Run the new vLLM compatibility tests
cd libs/partners/openai
uv run --group test pytest tests/unit_tests/chat_models/test_vllm_compatibility.py -v

# All 6 tests pass ✅

Code Quality

• ✅ Linting: All code passes ruff format and ruff check
• ✅ Type Safety: Proper error handling with try-catch blocks
• ✅ Documentation: Comprehensive docstrings and inline comments
• ✅ Error Messages: Maintains original error messages for debugging

Backward Compatibility

• ✅ No Breaking Changes: All existing functionality preserved
• ✅ Error Behavior: Original errors still raised when appropriate
• ✅ API Compatibility: No changes to public interfaces
• ✅ Performance: Minimal overhead (only executes fallback when needed)

Related Issues

• Fixes LangChain-OpenAI raises error due to null choices when using vLLM OpenAI-compatible API langchain-ai/langchain#32252
• Related to vLLM integration issues with OpenAI-compatible APIs

Review Checklist

• Stable Public Interfaces: No breaking changes to exported APIs
• Type Hints: All functions have complete type annotations
• Testing: Comprehensive unit tests covering happy path and edge cases
• Security: No dangerous patterns, proper exception handling
• Documentation: Google-style docstrings with clear explanations
• Code Quality: Passes lint and format checks
• Architecture: Clean, maintainable solution with proper error handling