← Runners

LlamaEdgeLlamaEdge

LlamaEdge 社区（WasmEdge / Second State 生态）LlamaEdge community (WasmEdge / Second State ecosystem)

本地推理Local inference 服务化推理Inference serving OpenAI兼容OpenAI-compatible WASMWASM WasmEdgeWasmEdge GGUFGGUF Apache-2.0Apache-2.0 端侧On-device partial lifecycle: maintenance 核验Checked 2026-10-31

LlamaEdge：WasmEdge 运行时上的 GGUF 推理 Wasm 应用集（Apache-2.0，1,663★）—— 「run LLM inference apps and create OpenAI-compatible API services locally」：装 WasmEdge → 下载 GGUF + llama-chat.wasm → 即跑；.wasm 应用跨 CPU/GPU 设备可移植是核心卖点。口径要点：主仓推送停 2026-02-08、release 0.29.0 同日——lifecycle=maintenance（文档站在线非停运）。

LlamaEdge: GGUF inference Wasm apps on the WasmEdge runtime (Apache-2.0, 1,663 stars) — "run LLM inference apps and create OpenAI-compatible API services locally": install WasmEdge → download a GGUF + llama-chat.wasm → run; the .wasm app's portability across CPU/GPU devices is the core selling point. Key wording: main-repo pushes stopped 2026-02-08, release 0.29.0 published the same day — lifecycle=maintenance (the docs site being online means not retired).

适合与不适合When to use it, when not to

适合：WASM 可移植路线、边缘/嵌入式设备；WasmEdge 生态用户。不适合：要活跃发版节奏；要 llama.cpp 原生全特性面。

Fits: the portable-WASM route, edge/embedded devices; WasmEdge-ecosystem users. Not a fit: an active release cadence; llama.cpp's full native feature surface.

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-31

出品属地Origin: 国外出品foreign-made　LlamaEdge 社区（Second State / WasmEdge 生态，CNCF 毕业项目 WasmEdge 同系）。LlamaEdge community (Second State / WasmEdge ecosystem, same lineage as the CNCF-graduated WasmEdge).
国内使用Direct access: 大陆可直连direct OK　llamaedge.com 直连 200；GitHub 仓直连 200（2026-10-31）。llamaedge.com and the GitHub repo both returned 200 direct (2026-10-31).
备案状态CAC filing: 未见于备案名单not in filing list：未见于网信办公示名单（2026-10-31 检索）；开源项目未面向大陆公众提供服务，备案不适用。not in the CAC public registry (searched 2026-10-31); an open-source project not serving the mainland public, so filing does not apply.

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.

固定坐标系Fixed coordinate system

8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.

模型与开放条件Model & openness: GGUF 本地权重 + WASM 应用层——模型走 GGML/AUTO 预加载（README 命令口径），推理逻辑编成 Wasm 应用（llama-chat.wasm 等）跑在 WasmEdge 运行时上；覆盖 LLM/embedding/语音转写/TTS/文生图/多模态六类模型线（README 分类链接）。
Local GGUF weights + a WASM app layer — models load via GGML/AUTO preloading (README command wording), while the inference logic is compiled into Wasm apps (llama-chat.wasm and friends) running on the WasmEdge runtime; six model lines covered: LLM/embedding/speech-to-text/TTS/text-to-image/multimodal (README category links).
运行位置Where it runs: WasmEdge 运行时上的跨平台 Wasm 应用——「cross-platform portable Wasm app that can run on many CPU and GPU devices」（README）；先装 WasmEdge，再下载 GGUF + .wasm 应用即跑。 GitHub LlamaEdge/LlamaEdge Apache-2.0 1,663★ 推送停 2026-02-08（API 核验）；最新 release 0.29.0 同日（2026-02-08）发布。
Cross-platform Wasm apps on the WasmEdge runtime — "cross-platform portable Wasm app that can run on many CPU and GPU devices" (README); install WasmEdge first, then download a GGUF + a .wasm app and run. GitHub LlamaEdge/LlamaEdge Apache-2.0 1,663 stars, pushes stopped 2026-02-08 (API verified); the latest release 0.29.0 was published the same day (2026-02-08).
本地文件Local files: 读写本机文件——GGUF 模型文件、.wasm 应用文件；wasmedge --dir .:. 挂载当前目录。
Reads and writes local files — GGUF model files, .wasm app files; wasmedge --dir .:. mounts the current directory.
关机后的任务Tasks after shutdown: CLI/本地服务进程（OpenAI 兼容 web 服务端点：/v1/chat/completions、/v1/embeddings、 /v1/audio/transcriptions 等——README 端点列表）；无常驻云面。
A CLI/local-service process (OpenAI-compatible web endpoints: /v1/chat/completions, /v1/embeddings, /v1/audio/transcriptions, etc. — README endpoint list); no persistent cloud surface.
工具与扩展Tools & extensions: OpenAI 兼容 API 端点面 + CLI chat 应用；不是 agent 工具宿主。
The OpenAI-compatible API endpoint surface + the CLI chat app; not an agent-tool host.
上下文与记忆Context & memory: 无跨会话记忆（推理服务面）——上下文长度随 GGUF 模型与 WasmEdge 预加载参数配置；对话状态由客户端携带，服务端不存会话。
No cross-session memory (it is an inference-serving surface) — context length follows the GGUF model and WasmEdge preload parameters; conversation state is carried by the client, the server stores no sessions.
权限与限制Permissions & limits: 本机运行；权限面=本机文件与 CPU/GPU。Wasm 沙箱是运行时层卖点（WasmEdge 口径）。
Runs locally; the permission surface = local files and CPU/GPU. The Wasm sandbox is the runtime layer's selling point (WasmEdge wording).
适合什么任务Best for: 适合：要 WASM 可移植路线——一份 .wasm 应用跨 CPU/GPU 设备；边缘/嵌入式场景（WasmEdge 生态）。不适合：要活跃发版节奏（主仓推送停 2026-02-08）；要 llama.cpp 原生全特性面（WASM 层是转译路线）。
Fits: the WASM portability route — one .wasm app across many CPU/GPU devices; edge/embedded scenarios (the WasmEdge ecosystem). Does not fit: an active release cadence (main-repo pushes stopped 2026-02-08); llama.cpp's full native feature surface (the WASM layer is a transpiled route).

头号误解Top misconception

按旧 org 名找仓——现役主仓 LlamaEdge/LlamaEdge（README 内仍引 second-state/LlamaEdge 下载链接，两 org 同生态）Looking for the repo under the old org name — the live main repo is LlamaEdge/LlamaEdge (the README still links second-state/LlamaEdge downloads; both orgs are the same ecosystem)
把推送停 2026-02-08 当停运——llamaedge.com 文档线在线（直连 200）、release 0.29.0 同日发布；lifecycle 按 maintenance 口径Reading the 2026-02-08 push stop as retirement — the llamaedge.com docs line is online (200 direct) and release 0.29.0 shipped the same day; lifecycle follows the maintenance wording
忘装 WasmEdge 前置——它不是独立二进制 runner，是 WasmEdge 上跑 .wasm 应用；install_v2.sh 是第一步Forgetting the WasmEdge prerequisite — it is not a standalone binary runner but .wasm apps on WasmEdge; install_v2.sh is step one
把六类模型线当同等成熟——LLM 线是主仓主线，embedding/语音/图像线在各自仓与文档页，成熟度不同Treating the six model lines as equally mature — the LLM line is the main repo's core; embedding/speech/image lines live in separate repos and docs pages with different maturity

未知项清单Known unknowns

停更后是否有迁移/继任口径（org 内多数仓 2025 年起停更，未见官方声明）
各 GPU 后端（CUDA/Metal）在 WasmEdge 侧的支持矩阵（文档有页，未逐条核验）
六类模型线各自的发布仓与版本节奏（README 分类链接指向不同仓）

Whether there is any migration/successor statement after the update pause (most repos in the org stopped updating from 2025; no official statement found)
The support matrix of each GPU backend (CUDA/Metal) on the WasmEdge side (docs pages exist; not verified item by item)
The release repos and version cadence of each of the six model lines (README category links point to different repos)

证据来源Evidence sources

判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-31，内容更新日content updated on 2026-10-31。

类型Type	名称Name	链接Link
repo	LlamaEdge/LlamaEdge · 仓库（Apache-2.0；1,663★，推送停 2026-02-08，GitHub API 核验；README 快速开始+端点列表）LlamaEdge/LlamaEdge · repository (Apache-2.0; 1,663 stars, pushes stopped 2026-02-08, GitHub API verified; README quick start + endpoint list)	https://github.com/LlamaEdge/LlamaEdge
changelog	GitHub releases · LlamaEdge（变更记录线：最新 release 0.29.0，2026-02-08 发布）GitHub releases · LlamaEdge (change-record line: latest release 0.29.0, published 2026-02-08)	https://github.com/LlamaEdge/LlamaEdge/releases
docs	LlamaEdge · 官方文档站（llamaedge.com/docs 直连 200；六类模型分类页）LlamaEdge · official docs site (llamaedge.com/docs returned 200 direct; six model-category pages)	https://llamaedge.com/docs

实测记录Measured runs

本档未跑吞吐与显存实测；结论只覆盖运行时能力与许可。

This site has not completed measured runs. The test protocol is at protocols/runner/_protocol.md.