ai-startup-skills
创业公司做 AI 产品的 7 步 Agent Skill 合集:验证 → BP → 政策 → 路演 → 获客 → 定价 → 注册。
Translation pending
Explore public MODs by lifecycle phase, serving scenario, platform, readiness, and evidence scope.
Translation pending
OtherPython · JavaScriptGitHub @vLLM-HUSTOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
Owners of public repositories under a personal GitHub account can sign in to verify ownership and manage the introduction themselves. No administrator approval is needed.
Sign in, generate a verification file, and commit it to this repository’s default branch within 30 minutes. Files on other branches or forks do not count. Then check the file below to manage the project without administrator approval. Keep the file to retain your access.
If needed, you will go to sign-in. After signing in, return here and press your chosen verification button again.
Run /nomorevibe verify in your project folder to verify ownership and manage the introduction yourself.
Verify in one minuteAsk about other verification options
vllm-hust official website and benchmark-driven serving showcase
Translation pending
Taken from the repository README
Loading the full README.
Read the full README ↗数据提交规则:本仓库不接受数据 PR。 实验结果、曲线、矩阵、基线、价格依据、元数据与证据 必须 PR 到 vllm-hust-benchmark。 网站只接收展示和代码变更。详见 贡献规则。
vllm-hust 项目概览 vllm-hust 围绕上游 vLLM 展开,聚焦国产算力适配、AGI4S 服务场景和 benchmark 驱动验证。
/v1/chat/completions 和流式响应请勿用下方 PyPI 包安装当前服务栈,它属于不同的版本组合;先查阅维护中的部署指南,选择经过验证的 core/plugin/镜像组合。
# 获取部署工具(并不直接安装运行时)
git clone https://github.com/vLLM-HUST/vllm-hust-dev-hub.git
该区块由 data/version_meta.json 驱动,运行 python scripts/sync_version_meta.py 自动更新。
data/version_meta.jsonpython scripts/refresh_site_status.py --refreshpython scripts/sync_version_meta.py.github/workflows/site-status-check.yml.github/workflows/sync-version-meta.yml (does not refresh source
SHAs)bash scripts/check_stale_versions.shdocs/VERSION_METADATA.md首页现在支持把远端 vllm-hust-workstation 作为一个可配置面板展示出来,适合把部署在远端算力节点上的 workstation 直接呈现到 website 界面。
data/workstation_embed.jsonassets/workstation-embed.jsembed(iframe 内嵌)或 link(仅提供打开入口)示例:
{
"enabled": true,
"mode": "embed",
"workstation_url": "https://ws.sage.org.ai",
"backend_url": "https://api.sage.org.ai",
"docs_url": "./docs/CLOUDFLARE_SETUP.md",
"label_zh": "vLLM-HUST 工作站",
"label_en": "vLLM-HUST Workstation"
}
注意:
website 是静态站点,不代理 workstation API;要展示远端控制台,需要 workstation 自己对外可访问。APP_FRAME_ANCESTORS。推荐的 sage.org.ai 域名拆分:
vllm-hust.sage.org.ai:GitHub Pages 静态官网ws.sage.org.ai:远端节点上的 vllm-hust-workstationapi.sage.org.ai:远端节点上的 vllm-hust OpenAI 兼容接口docs/ARCHITECTURE_TASK_MAPPING.mddocs/IMPLEMENTATION_PRIORITIES.mdwebsite does not ingest raw compare directories and does not hand-edit benchmark rows.
The dataset validation results surface is a separate, read-only path. The independent fixed-machine
validation service runs after a PR, then publishes a versioned result artifact for the website to
display at dataset-validation.html. The website does not trigger that
service and does not perform evaluation, admission, publication, or merge-gate work. The frontend
contract and empty local fixture are documented in
data/DATASET_VALIDATION_RESULTS.md. The existing leaderboard
snapshot chain below remains a legacy compatibility surface until its replacement data contract is
explicitly migrated; it must not be used as the dataset validation service contract.
The only supported leaderboard data chain is:
leaderboard_manifest.json +
*_leaderboard.json)vllm-hust-benchmark/leaderboard-data/snapshotsintellistream/vllm-hust-benchmark-resultsdata/leaderboard_single.jsondata/leaderboard_multi.jsondata/leaderboard_historical.json (recovered trend-only records)data/leaderboard_compare.jsondata/last_updated.jsonLeaderboard rendering consumes benchmark GitHub raw snapshots first, then HF, then the checked-in
website projection. leaderboard_compare.json provides neutral engine-vs-engine head-to-head views.
Leaderboard version rendering follows a split UI contract:
docs/VERSION_METADATA.md#leaderboard-version-display-contractleaderboard_compare.json now also carries a mandatory hard-constraint snapshot derived from
validated entries:
All hard-constraint records must declare constraints.scenario_source = vllm-benchmark, using the
vLLM benchmark inheritance path as the accountable scenario and dataset source.
本仓库提供统一的 quickstart.sh 模式入口:
--standard:标准模式,依赖优先从 PyPI 安装(稳定/发布导向)--dev:开发模式,在 standard 基础上尝试本地 editable 覆盖(使用 --no-deps)requirements-dev.txt:本地校验依赖单一来源,供 quickstart、CI 和 scripts/validate-local.sh 统一复用vllm-hust-website)# 查看帮助
./quickstart.sh --help
# 标准模式
./quickstart.sh --standard
# 开发模式(默认)
./quickstart.sh --dev
# 可选:跳过清理或 hooks
./quickstart.sh --dev --skip-cleanup --skip-hooks
注意:脚本不会创建
venv/.venv,请使用当前已配置的非-venv Python 环境。
本地校验依赖以 requirements-dev.txt 为唯一声明源。首次进入仓库后,执行下面任一方式即可获得和 CI 一致的基础 Python 工具链:
# 推荐:同时安装校验依赖并配置 hooks
./quickstart.sh --dev
# 或者仅安装校验依赖
python3.11 -m pip install -r requirements-dev.txt
在本地提交前,如果你想尽量一次性对齐 CI 的 validate job,直接运行:
./scripts/validate-local.sh
这个脚本会依次执行:
pre-commit run --all-filespytest tests/ -vquickstart.sh 安装语句校验要求:
python3.11pre-commit / pytestuv,脚本会自动使用 uv run --python 3.11 --with-requirements requirements-dev.txt ...
兜底,继续复用同一份依赖声明如果要把静态站点在本机以 systemd --user 常驻方式挂起来,而不是手动执行 python3 -m http.server 8000,仓库现在内置了最小管理脚本:
# 一次性安装 / 更新 service
./scripts/deploy_website_service.sh install-service
# 安装并立即重启到最新配置
./scripts/deploy_website_service.sh deploy
# 查看状态 / 日志
./scripts/deploy_website_service.sh status
./scripts/deploy_website_service.sh logs
默认行为:
127.0.0.18000vllm-hust-website可选环境变量:
export WEBSITE_HOST=127.0.0.1
export WEBSITE_PORT=8000
export WEBSITE_ROOT_DIR=/path/to/vllm-hust-website
export WEBSITE_SYSTEMD_SERVICE_NAME=vllm-hust-website