Open Source

Training-free acceleration for video DiTs by caching and reusing block features across denoising steps.

Linear multistep feature forecasting that adaptively skips denoising steps; supports FLUX, DiT, and HunyuanVideo.

Accelerating autoregressive video generation with cache, predict, and recompute modes and active-frame selective computation.

Edge-collaborative AIGC task scheduling with attention-guided diffusion reinforcement learning.

A layer-aware, resource-adaptive container scheduler built on the Kubernetes scheduling framework.

A research starter handbook for undergraduates in AI systems, with methods, fundamentals, environment setup, and 9 hands-on labs (in Chinese).

Nineteen open-source Agent Skills for CS researchers using Claude Code, Codex, and other coding agents, covering LaTeX linting, citation verification, scheduling simulators, LLM serving benchmarks, result statistics, paper figures, and pre-submission checks (in Chinese).

Over sixty browser-based tools for writing papers, running experiments, and managing servers, including citation verification and formatting (GB/T 7714, APA, IEEE), LaTeX and PDF submission checks, statistical tests and plotting, GPU memory and training-time estimation, CCF deadline countdowns, and Slurm/Kubernetes config generators. Everything runs locally in the browser (in Chinese).

Chinese Deep Learning course site based on Qiu Xipeng's Neural Networks and Deep Learning, with lecture notes, slides, exercises, a bilingual glossary, and interactive demos for activations, optimizers, and convolutions.

Chinese reinforcement learning tutorials from MDP, Bellman equations, and Q-learning to DQN, PPO, SAC, offline RL, and LLM RLHF / DPO / GRPO, with runnable PyTorch code, paper digests, and browser interactive tools.