Sophon LLMOps provides corpus full life cycle management capabilities such as large-scale corpus access, guided corpus cleaning, multi-team collaborative corpus annotation, corpus evaluation and operation; in the knowledge storage stage, the platform supports multiple file format parsing and vectorization, and supports professional parsing, sharding and retrieval output strategy configuration: supports users to publish the parsing process arranged by the application development module into a parsing service, and realizes more flexible document content extraction. In addition to self-built knowledge bases, the platform can also support docking with existing knowledge bases within the enterprise.

Sophon LLMOps supports unified management of multimodal (LLMs) models and third-party model services, unified interface specifications, and refined version management of different modal models. In terms of model training, the platform provides two LLM fine-tuning training methods: interface guidance and code programming. Based on the platform's model evaluation function, you can choose preset general evaluation rules or custom evaluation templates to evaluate indicators in various dimensions for the models that have been put on the shelves. In particular, Sophon LLMOps is still compatible with the full life cycle capabilities of training and management of traditional machine learning and deep learning models.

Sophon LLMOps provides four application building methods for different user types and usage scenarios: zero-code building similar to GPTs, low-code operator orchestration, and building LLM applications based on online programming and custom containers. For example, for operators, the platform supports users to quickly build RAG and Agent applications in a zero-code way. Users can publish the built application with one click for quick experience, share it with one click, and provide API calls to the outside world.

Sophon LLMOps supports configuring elastic scaling strategies and dynamically scaling capacity to achieve load balancing. It also supports multi-version traffic distribution strategies and has grayscale release and A/B testing capabilities. During the monitoring and deployment phase, the platform supports real-time monitoring of the operating status of model services and controls model production environment risks, such as cluster resource utilization, throughput, response time, and access records.

For enterprise organizations, Sophon LLMOps provides a series of enterprise-level capabilities including member permission control, resource monitoring, billing management, approval and auditing, security control, etc. Member management and permission control within the space ensure data isolation and data security between users, support users to effectively audit and approve the core processes of the platform such as model deployment, application startup and shutdown, and task operation, and enhance the traceability of operations and the security of the system.