vLLM-Kunlun (Baidu Kunlun Chip LLM Inference Framework)

LLM inference framework for Baidu Kunlun XPU that adapts capabilities such as PagedAttention to Kunlun P800, enabling high-performance inference for mainstream models such as Qwen.

SGLang

High-performance LLM serving and inference framework focused on optimized runtime and multi-backend deployment.

RLLM

Research framework for reinforcement learning and LLM training workflows with integrations such as verl.

vLLM-Omni

Omni and multimodal inference extension built on vLLM for speech and text generation scenarios.

OpenRLHF

Distributed training framework for RLHF and alignment experiments.

Transformers

Flagship Hugging Face library for model implementations plus training and inference utilities.

Haikou Police Administrative Knowledge Base

RAG System Construction and LLM Evaluation (Algorithm)

Intelligent administrative knowledge base for the Haikou Police system, using RAG to improve answer quality for administrative knowledge queries.