Model Support

---

Scope of Model Support

To ensure the stability and maintainability of the SERVICEME platform in multi-cloud and multi-model environments, we have established unified support specifications for the integrated large language models (LLM), vector models (Embedding), speech and document recognition models.

These specifications aim to:

• Reduce the complexity of multi-model compatibility and operations
• Improve overall performance and controllability
• Ensure stable availability of new features on mainstream models

Model Support Levels

Standard Support

• Refers to models that have been fully validated and incorporated into the product's standard testing process.
• These models have passed comprehensive tests in functionality, performance, and stability, and provide service availability guarantees (SLA).

Experimental Support

• Refers to models that have integration capabilities and can be used in the main workflow.
• Only basic functionality tests have been completed (interface connectivity, main workflow verification), with no guarantee of long-term performance or stability.
• Clearly marked as "Experimental Support" in product documentation.

Model Categories and Support Status

Model Type	Example Models	Current Support Level	Remarks
Large Language Models (LLM)	GPT-4, GPT-4o, Claude 3, Gemini 1.5	Standard Support	Completed multi-round testing and integration
Vector Models (Embedding)	text-embedding-3-small / large	Standard Support	Suitable for knowledge retrieval and semantic matching
Text-to-Speech (TTS)	Azure Speech, OpenAI TTS	Experimental Support	Functionality available, performance under verification
Image Recognition / OCR	Azure Vision, Baidu OCR	Standard Support	Limited to text recognition scenarios
Speech-to-Text (STT)	Azure Speech-to-Text	Experimental Support	Planned to be included in standard support later

Integration and Verification Requirements

1. Unified Interface Specification

   • All models must be called through a unified Model Gateway service.
   • Request and response formats must comply with the SERVICEME model interface specification.

2. Security and Compliance

   • Calling third-party public APIs without security verification is prohibited.
   • Integrated models must comply with company data compliance requirements (privacy, security, regional restrictions, etc.).

3. Performance and Observability

   • Models must support call logging and performance monitoring.
   • Health checks and circuit breaker mechanisms must be established for standard support models.

Future Plans

• Continuously expand the standard support scope for mainstream models (such as Mistral, Yi, Qwen series).
• Provide model performance benchmark testing and compatibility reports.
• Launch a model integration certification mechanism (Model Certification).