Global

8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.

Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦

8B to 27B: a calibrated probability for every option in one pass, and a short thought only when unsure. cpp and MLX.

A high-throughput and memory-efficient inference and serving engine for LLMs

1-Flash-1017 (748B MoE) privately on your own GPU(s), from one card to 16, NVIDIA or AMD, Linux or Windows. Engine, server, dashboard, OpenAI- and Anthropic-compatible API, quant tools.

Implementing LLM architectures from scratch in PyTorch

预览地址为 ，简体中文位于根目录，英文与繁體中文分别位于 en/ 和 zh-tw/。生成文件位于 build/，详细说明见网站构建与发布。

Build a modern LLM from scratch. Every line commented. Explained like we are five.

NOTE] > The total size of DeepSeek-V3 models on Hugging Face is 685B, which includes 671B of the Main Model weights and 14B of the Multi-Token Prediction (MTP) Module weights.

24 GB of RAM. Portable C99: no BLAS, no framework, no GPU.

The scripts load VPP2 directly from this checkout; installing the vpp2 package is unnecessary. txt.

The fastest way to run Qwen3.8-Flash-Next on Strix Halo (gfx1151)

An open-source AI accelerator, developed by AI: RTL, ISA, simulator, compiler and profiler in one repo. 5 on a Kintex-7 PCIe card.

DeepGEMM is a unified, high-performance tensor core kernel library that brings together the key computation primitives of modern large language models — GEMMs (FP8, FP4, BF16), fused MoE with overlapped communication (Mega MoE), MQA scoring for the lightning indexer…

Kimi K3 is an open-weight, native multimodal agentic model and our most capable model to date. 8T-parameter model built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), with native vision capabilities and a 1-million-token context window.

主要记录大语言大模型（LLMs） 算法（应用）工程师相关的知识及面试题

📍 Use GLM-5.3 & GLM-5.3-Flash API services on Z.ai API Platform.

8-27B bf16 (130 tok/s), 262k context, prefix caching. Works with Claude Code, Codex, opencode and pi.

DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.

Qwen3 is the large language model series developed by Qwen team, Alibaba Cloud.

8 is the large language model series developed by Qwen team, Alibaba Group.

AI System Design - Learn how to design AI systems built on LLMs, RAG, and AI Agents step by step.

A Flexible Framework for Experiencing Heterogeneous LLM Inference/Fine-tune Optimizations

3-Flash beyond available RAM by streaming activated weights directly from NVMe. A dependency-free, embeddable C inference engine.

Hy-MT2 is a family of “fast-thinking” multilingual translation models designed for complex real-world scenarios. 8B, 7B, and 30B-A3B (MoE), all of which support translation among 33 languages and effectively follow translation instructions in multiple languages. 5x.

\janas-chat. 8 times as fast with the dense model (Qwen3-4B 346 tok/s against 193), whose whole blocks go to the GPU, through Intel's own Windows driver there.

A theoretical reconstruction of the Claude Mythos architecture, built from first principles using the available research literature.

DeepGEMM-Ascend: clean and efficient matrix multiplication kernel library for Huawei Ascend NPUs

Make sure to download the checkpoint and place the ckpt-0 directory in checkpoints - see Downloading the weights

DeepEP (DeepEveryParallel) is a high-performance communication library for modern machine learning training and inference.