Sglang is a high-performance inference engine and serving system designed for large language and multimodal models. It provides a programmable interface for orchestrating complex generation workflows, enabling developers to coordinate multi-turn dialogues, tool invocations, and reasoning chains through a domain-specific language. The platform is built to support production-scale deployments, offering an OpenAI-compatible API that allows for integration with existing application ecosystems. The system distinguishes itself through a disaggregated architecture that separates compute-intensive pr
FlashInfer is a library of high-performance GPU kernels purpose-built for accelerating large language model inference. It provides optimized implementations for attention operations (including flash attention, page attention, multi-head latent attention, and cascade attention) using paged key-value caches, fused kernel composition, and just-in-time compilation. The library also includes specialized kernels for mixture-of-experts layers, block-scaled low-precision quantization (FP8, FP4), and distributed collective communication. What distinguishes FlashInfer is its fused all-reduce communicat
Tiny Universe is an educational monorepo that delivers multiple independent implementations of core AI subsystems as self-contained Jupyter notebooks. It provides from-scratch constructions of foundational architectures including a complete Transformer model built from the original paper specification, a denoising diffusion probabilistic model for image generation, and a ReAct-style autonomous agent framework that equips an LLM with tools for planning and multi-step task execution. The project distinguishes itself by covering the full lifecycle of modern AI systems through hands-on implementa
This project is a manual reconstruction of the Llama 3 transformer architecture implemented as a PyTorch neural network. It serves as a reference for the internal mathematical structure and tensor flow of a transformer-based language model designed for next token prediction. The implementation focuses on building the model from scratch using basic matrix operations and tensor manipulations. It demonstrates the manual construction of core components, including rotary positional embeddings, multi-head self-attention, and root mean square normalization. The codebase covers the full inference pi