Xinyuan-LLM-14B-0428 Overview

Xinyuan-LLM-14B-0428, developed by Cylingo Group, is a 14-billion parameter foundational language model based on the Qwen3-14B architecture. It stands out as the first foundational model specifically designed for the mental health industry, having been extensively fine-tuned on millions of diverse mental health-related data points.

Key Capabilities

• 24/7 Intelligent Mental Health Support: Provides continuous, round-the-clock assistance.
• Diverse Scenario Coverage: Addresses a wide range of mental health situations and challenges.
• Personalized Psychological Profiling: Capable of building individualized psychological profiles for users.
• Customized Family Companion Solutions: Offers tailored solutions for parenting and family-related mental health issues.

Deployment and Usage

This model supports deployment via sglang (version >=0.4.6.post1) or vllm (version >=0.8.5) to create OpenAI-compatible API endpoints. It is also compatible with local applications such as Ollama, LMStudio, MLX-LM, llama.cpp, and KTransformers. For optimal performance in non-thinking mode, recommended inference parameters include Temperature=0.8, TopP=0.8, TopK=20, and MinP=0. Users are advised to add rope_scaling configuration only when processing long contexts, adjusting the factor as needed for typical application context lengths.