AI Infra2025
Provides high-throughput, low-latency GPU communication kernels for Mixture-of-Experts (MoE) and expert-parallel workloads, with NVLink↔RDMA-aware forwarding, FP8/BF16 support, and low-latency RDMA hooks for inference decoding.
AI Infra2025
Optimized MLA (Multi-head Latent Attention) decoding kernels powering DeepSeek-V3/V3.2 inference on Hopper and Blackwell GPUs. Dense decoding reaches ~3000 GB/s and 660 TFLOPS on H800; the sparse path stores the KV cache in FP8.
AI Infra2025
Runs and fine-tunes LLMs locally on Apple silicon via the MLX framework, pulling thousands of Hugging Face models with one command. Adds 4- and 8-bit quantization, LoRA and full fine-tuning, prompt caching, and distributed inference across Macs.
AI Deploy2025
Splits LLM inference into separate prefill and decode GPU pools, then routes requests with KV-cache awareness to cut redundant recomputation. Reports up to 30x throughput on DeepSeek-R1 (GB200 NVL72) and works across TensorRT-LLM, vLLM, and SGLang.
AI Deploy2025
Run large-language and multimodal models locally on edge devices (Android, iOS, desktop, web, Raspberry Pi) with hardware acceleration, function-calling, and multi-language SDKs—designed for low-latency, privacy-sensitive on-device inference.
AI Infra2025
Measures generative AI inference performance with token-level metrics (TTFT, inter-token latency), latency, and throughput under realistic traffic patterns. Provides a multiprocess engine, real-time TUI dashboard, extensible plugins, and integrations for telemetry and result uploads, aimed at inference benchmarking and capacity planning.
MCP Client2025
Official Go implementation of the Model Context Protocol for building MCP servers and clients. Tool handlers are type-safe, with JSON schemas inferred from Go structs via generics. Ships stdio, command, streamable-HTTP, SSE, and in-memory transports.
AI Image2025
Converts document images—scans, photos, born-digital PDFs—into structured text in two stages: first map layout and reading order, then parse each element (text, tables, formulas, figures) in parallel, each guided by its own task prompt.
AI Infra2025
Reimplements the vLLM inference engine from scratch in ~1,200 lines of readable Python, matching its offline throughput on small models. Prefix caching, tensor parallelism, torch.compile, and CUDA graphs are all kept legible.
AI Agent2025
Unifies agentic tasks, reasoning, and coding in a single MoE model with 355B total / 32B active parameters and a switchable thinking mode. A lighter 106B-param Air variant trades scale for efficiency; both ship MIT-licensed.
AI Infra2025
An open-source memory layer that turns agent runs and conversations into structured, persistent state recallable across sessions. Captures facts, events, preferences, and relationships automatically; LLM-agnostic with SDK and MCP integration.
AI Agent2025
A TypeScript agent harness split into composable npm packages: a unified LLM API across OpenAI, Anthropic and Google, an agent runtime with tool calling and state, a self-extensible coding-agent CLI, and a differential-rendering terminal UI library.