dcu

国产加速卡-海光DCU实战（大模型训练、微调、推理 等）

K100AI 模型部署实践 —— 海光 K100-AI 上大模型的成功配置线拉起包（当前：Qwen3.8-27B 六条精选线；持续更新）

Semester 1 work

Scrape's all information for a course from the DCU timetable website.

海光 K100-AI（gfx928）上跑通并优化 Qwen3.8-Flash-Next INT8 W8A8：4 卡 TP4，单流 31 tok/s、20 并发聚合 175 tok/s，含 262K 上下文、MTP 投机解码、工具调用与完整实测

comsci1/ca116/computer-programming-1

A reusable Codex skill and engineering playbook for porting CFD/PDE and scientific-computing codes to Hygon DCU/DTK/HIP.

📚 Full-stack web dev labs — Python, Django, JavaScript, HTML/CSS

comsci1/ca117/computer-programming-2

This is a python script for a very specific scenario in DCU's CSC1004 Computer Programming II.

🍕 Pizza ordering web app — built with Django & Python

Semester 2 work

多厂商 vLLM 推理全流程探测工具箱：HTTP→调度→预填充→解码→投机解码→GPU kernel 逐层观测，中文报告 + 标准化对比套件（NVIDIA/AMD/Hygon/Intel/Ascend）

🎮 Game store e-commerce app — Django, Python, HTML/CSS

Compatibility validation engine for domestic processors: turns chip models plus workload scenarios into a compatibility matrix with an evidence chain, memory sizing, tuning advice and a traceable FAQ. Detects conflicting public sources instead of silently picking one. 信创适配性验证引擎：芯片型号 + 业务场景进，兼容性矩阵、调优建议、排查FAQ 出。

Meta-skill for integrating HPC, AI4S and scientific applications with SCNet compute resources, Slurm scheduling and cluster packaging, including Hygon DCU compatibility.

A miniature of vLLM v1 with a vLLM-style Platform/op-family hardware abstraction layer. One engine, four backends: NVIDIA CUDA, Hygon DCU (DTK/HIP), Ascend NPU, Moore Threads MUSA. Real continuous batching, paged KV cache, TP, CUDA graph — runs Qwen3/Olmoe end-to-end

Agent skill for working on SCNet clusters

📖 Library management system — Django web app with Python