AI Deploy2021
Serves predictive and generative ML models on Kubernetes via a single InferenceService CRD, with scale-to-zero, canary rollouts, and an OpenAI-compatible LLM path on vLLM. One autoscaling abstraction over PyTorch, XGBoost, ONNX, and HuggingFace.
AI Infra2021
Drop-in transformer building blocks with custom CUDA kernels: memory-efficient exact attention (up to ~10x faster), block-sparse attention, fused softmax/layernorm/SwiGLU ops. Cuts VRAM and speeds up diffusion and LLM training on Nvidia GPUs.
AI Train2021
HPC-AI Technology Inc. (Colossal-AI team), Shenggui Li +1HPC-AI Technology Inc., National University of Singapore
Scales a single-GPU training script to thousands of GPUs through a unified interface, combining data, pipeline, tensor, and sequence parallelism. Its Gemini memory manager offloads tensors across GPU, CPU, and NVMe so models far larger than VRAM still fit.
AI Infra2022
A continuously-updated, categorized self-hosting guide that catalogs tools, deployment notes and resources for containers, networking, home automation and running LLMs/chatbots locally; provided as a long, structured README with links and quick setup tips.
AI Infra2022
Compiles plain Python functions into GPU or CPU kernels at runtime via a JIT decorator, with differentiable output that plugs into PyTorch, JAX, and Paddle. Ships physics, robotics, geometry, and FEM primitives — particles, meshes, ray-casting, FFT.
AI Infra2022
Provides runtime support for reversible effects and reactive coeffects so components can be declared, composed, and hot-replaced safely; includes effect tracking, coeffect resolution, a declarative component loader and HMR — aimed at plugin-driven agent harnesses and dynamic systems.
AI Infra2022
Fused CUDA kernels that compute exact attention without ever writing the full N×N score matrix to GPU memory, cutting memory from quadratic to linear and speeding up training and inference on A100/H100. Ships FlashAttention-2/3 plus KV-cache decode paths.
AI Infra2022
Collects metrics, distributed traces, and continuous profiles via eBPF with zero code instrumentation, covering apps in any language plus gateways, service meshes, databases, and queues. Profiling adds under 1% overhead.
AI Infra2022
Smart model router for personal AI agents that sends each request to the cheapest model capable of handling it — cutting API costs by up to ~70%. Uses a fast 23-dimension scorer, automatic fallbacks, per-tier controls, and supports local Docker self-hosting or a cloud app; ideal for cost-sensitive personal agents.
AI Agent2022
Build, run, and monitor LLM agents across one stack: an open framework for chaining models and tools, LangGraph for stateful agent orchestration, and LangSmith for tracing, evaluation, and deployment in production.
AI Infra2022
Build full‑stack web apps entirely in Python — write frontend components and backend state as Python classes with a reactive model. Provides fast refresh, deployment tooling, and AI-focused integrations such as an AI Builder and an Agent Toolkit for connecting LLMs and image models.
AI Agent2022
Connects LLMs to private and domain-specific data with ingestion, indexing, and retrieval primitives for RAG and agentic apps. Centers on document parsing via LlamaParse for 90+ file formats, schema-based extraction, and composable queries.