[feat] Add CentML as llm provider

[V2arK]

Title

Add CentML provider support with granular function calling and JSON schema controls

Relevant issues

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• I have added a screenshot of my new test passing locally
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

Local test screenshot

Type

🆕 New Feature

Changes

Core Provider Integration

• Added CentML provider support to LiteLLM's provider ecosystem
• Added CENTML = "centml" to LlmProviders enum
• Integrated CentML into provider detection logic and OpenAI-compatible handler routing
• Added environment variable support (CENTML_API_KEY, CENTML_API_BASE)

Model Support & Capabilities Matrix

• Added 12 CentML models with differentiated capability support:
  • JSON ✅ / Tool ✅: Llama-3.3-70B-Instruct, Llama-4-Scout-17B-16E-Instruct, Llama-4-Maverick-17B-128E-Instruct-FP8, Llama-3.2-3B-Instruct, Llama-3.2-11B-Vision-Instruct
  • JSON ✅ / Tool ❌: DeepSeek-R1, DeepSeek-V3-0324, Phi-4-mini-instruct, Qwen2.5-VL-7B-Instruct
  • JSON ❌ / Tool ❌: QwQ-32B, Llama-Guard-4-12B
• Updated model pricing configuration in both main and backup JSON files

Granular Parameter Control

• Separated function calling and JSON schema controls into independent parameters:
  • supports_function_calling: Controls tools, tool_choice, function_call parameters
  • supports_response_schema: Controls response_format parameter independently
• Implemented CentmlConfig class with intelligent parameter validation based on model capabilities

Handler Implementation

• Created OpenAI-compatible handlers for both chat and text completion endpoints
• CentmlChatCompletion and CentmlTextCompletion classes extending OpenAI handlers
• Added transformation logic in CentmlConfig and CentmlTextCompletionConfig

Comprehensive Testing

• Added 20+ unit tests in tests/test_litellm/llms/centml/:
  • Chat completion transformation tests (20 tests)
  • Text completion transformation tests (18 tests)
• Added integration tests in tests/local_testing/test_centml_integration.py:
  • Parameter validation matrix testing
  • CentML documentation example verification
  • Error handling and authentication testing
• Added warning suppression for LiteLLM's Pydantic deprecation warnings

Technical Implementation Details

• Fixed model name lookup issues by properly handling centml/ prefix construction
• Added proper error handling for missing model capabilities
• Implemented response_format filtering (removes {"type": "text"} default)
• Added support for all standard OpenAI parameters where applicable

This implementation enables developers to use CentML's inference platform through LiteLLM with full parameter validation, ensuring that unsupported features are properly blocked while supported features work seamlessly with the existing LiteLLM ecosystem.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
litellm	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Jun 19, 2025 3:19pm

[CLAassistant]

All committers have signed the CLA.