[{"data":1,"prerenderedAt":192},["ShallowReactive",2],{"repo-eugr-spark-vllm-docker-en":3,"i18n-eugr-spark-vllm-docker-en":189},{"mirrorPool":4,"detail":13,"relatedProjects":74},[5,9],{"key":6,"label":7,"base":8,"type":6},"github","mirror.github","https:\u002F\u002Fgithub.com\u002F",{"key":10,"label":11,"base":12,"type":10},"auto-proxy","mirror.autoProxy","https:\u002F\u002Fgithub.xxlab.tech\u002F",{"owner":14,"repo":15,"repoId":16,"repoUrl":17,"starCount":18,"language":19,"description":20,"languages":21,"screenshots":26,"homepage":27,"pushedAt":28,"ownerAvatar":29,"ownerName":27,"ownerBio":27,"lastSyncTime":30,"lastReleaseTime":31,"tags":32,"license":33,"category":34,"subgroup":35,"releases":36},"eugr","spark-vllm-docker",1104120047,"https:\u002F\u002Fgithub.com\u002Feugr\u002Fspark-vllm-docker",2346,"Python","Docker configuration for running VLLM on dual DGX Sparks",{"Python":22,"Shell":23,"Dockerfile":24,"Jinja":25},50.2,41.7,4.4,3.6,[],"","2026-10-04","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F2192184?v=4","2026-10-04T13:15:39.582055+00:00","2026-10-04T12:54:03Z",[],"MIT","development","LLM推理\u002F微调工具",[37,48],{"tag":38,"publishAt":31,"semVer":39,"prerelease":41,"assets":42},"prebuilt-vllm-current",[40,40,40],0,false,[43],{"fileName":44,"sizeMb":45,"sha256":46,"downloadCount":40,"originUrl":47},"vllm-0.30.1rc1.dev631+g0872ddf4b.d20261004-cp312-cp312-linux_aarch64.whl",487.08,"00b4a676eefc45e9b9bd2d98278d7797f82734b9f25662c9e54487c3f5369ddd","https:\u002F\u002Fgithub.com\u002Feugr\u002Fspark-vllm-docker\u002Freleases\u002Fdownload\u002Fprebuilt-vllm-current\u002Fvllm-0.30.1rc1.dev631%2Bg0872ddf4b.d20261004-cp312-cp312-linux_aarch64.whl",{"tag":49,"publishAt":50,"semVer":51,"prerelease":41,"assets":52},"prebuilt-flashinfer-current","2026-10-04T12:47:18Z",[40,40,40],[53,59,64,70],{"fileName":54,"sizeMb":55,"sha256":56,"downloadCount":57,"originUrl":58},"flashinfer_jit_cache_sm121a-0.7.1-cp39-abi3-linux_aarch64.whl",175.76,"8e2e17dc335f0db91f54fede08ad38424b6467ca84ebab4d8c39189757dfb390",1,"https:\u002F\u002Fgithub.com\u002Feugr\u002Fspark-vllm-docker\u002Freleases\u002Fdownload\u002Fprebuilt-flashinfer-current\u002Fflashinfer_jit_cache_sm121a-0.7.1-cp39-abi3-linux_aarch64.whl",{"fileName":60,"sizeMb":61,"sha256":62,"downloadCount":40,"originUrl":63},"flashinfer_cubin-0.7.1-py3-none-any.whl",1479.49,"8a2d274c3179631c7a72176631c2383043755f7a79a3af864e41bb4af247b0af","https:\u002F\u002Fgithub.com\u002Feugr\u002Fspark-vllm-docker\u002Freleases\u002Fdownload\u002Fprebuilt-flashinfer-current\u002Fflashinfer_cubin-0.7.1-py3-none-any.whl",{"fileName":65,"sizeMb":66,"sha256":67,"downloadCount":68,"originUrl":69},"flashinfer_python-0.7.1-py3-none-any.whl",54.52,"97a43a0e57d8cc05ebcb3723a12ecf76bcfb04f246f46ae1f4e91054ea8ebcf4",2,"https:\u002F\u002Fgithub.com\u002Feugr\u002Fspark-vllm-docker\u002Freleases\u002Fdownload\u002Fprebuilt-flashinfer-current\u002Fflashinfer_python-0.7.1-py3-none-any.whl",{"fileName":71,"sizeMb":40,"sha256":72,"downloadCount":57,"originUrl":73},"flashinfer_jit_cache-0.7.1-cp39-abi3-manylinux_2_28_aarch64.whl","bb34cf3be0c45797489899861f825e62e94fcd50bfd6754420ac888473f9ada6","https:\u002F\u002Fgithub.com\u002Feugr\u002Fspark-vllm-docker\u002Freleases\u002Fdownload\u002Fprebuilt-flashinfer-current\u002Fflashinfer_jit_cache-0.7.1-cp39-abi3-manylinux_2_28_aarch64.whl",[75,94,127,147,159,175],{"owner":76,"repo":77,"repoId":78,"category":79,"subgroup":35,"language":80,"tags":81,"starCount":83,"ownerUid":84,"latestTag":85,"platforms":86,"lastReleaseTime":91,"downloadCount":92,"description":93},"ggml-org","llama.cpp",612354784,"ai","C++",[82],"ggml",131695,134263123,"b11734",[87,88,89,90],"windows","mac","linux","other","2026-10-19T14:08:40Z",481,"A C\u002FC++ implementation for LLM and VLM inference with minimal setup and state-of-the-art performance, supporting multiple model formats with CLI, server, and built-in Web UI.",{"owner":95,"repo":96,"repoId":97,"category":79,"subgroup":35,"language":98,"tags":99,"starCount":120,"ownerUid":121,"latestTag":122,"platforms":123,"lastReleaseTime":124,"downloadCount":125,"description":126},"diegosouzapw","OmniRoute",1157102282,"TypeScript",[100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119],"a2a","ai-agents","ai-gateway","anthropic","claude","claude-code","cline","codex","copilot","cursor","deepseek","free-ai","gemini","kimi","llm-gateway","mcp","openai","openai-proxy","qwen","token-saver",72815,8016841,"v3.8.51",[87,88,89,90],"2026-09-30T01:41:50Z",4681,"A free MIT AI gateway providing a unified endpoint to 290+ providers and 500+ models with quota-aware auto-fallback and token compression.",{"owner":128,"repo":129,"repoId":130,"category":131,"subgroup":35,"language":132,"tags":133,"starCount":140,"ownerUid":141,"latestTag":142,"platforms":143,"lastReleaseTime":144,"downloadCount":145,"description":146},"AlexsJones","llmfit",1158588224,"tools","Rust",[134,135,136,137,138,139],"gguf","llm","localai","mlx","skill","unsloth",37743,1235925,"v1.1.17",[87,88,89],"2026-10-08T18:04:05Z",1447,"A command-line tool that recommends the best local LLMs for your hardware configuration, supporting multi-GPU, MoE architectures, and multiple quantization formats.",{"owner":148,"repo":149,"repoId":150,"category":79,"subgroup":35,"language":98,"tags":151,"starCount":152,"ownerUid":153,"latestTag":154,"platforms":155,"lastReleaseTime":156,"downloadCount":157,"description":158},"tashfeenahmed","freellmapi",1217040185,[],37411,9307356,"v0.16.5",[87,88,89,90],"2026-10-19T11:30:09Z",1332,"A unified API gateway aggregating 34 free LLM providers and 635 free model endpoints, offering approximately 7.4 billion free tokens per month with smart routing, automatic failover, and encrypted key storage.",{"owner":160,"repo":161,"repoId":162,"category":79,"subgroup":35,"language":19,"tags":163,"starCount":168,"ownerUid":169,"latestTag":170,"platforms":171,"lastReleaseTime":172,"downloadCount":173,"description":174},"jundot","omlx",1157171418,[164,165,135,166,137,167],"apple-silicon","inference-server","macos","openai-api",22858,64250138,"v0.7.1rc1",[88,90],"2026-10-15T18:23:10Z",2195,"An LLM inference server optimized for Apple Silicon with continuous batching and SSD caching, managed directly from the macOS menu bar.",{"owner":176,"repo":177,"repoId":178,"category":79,"subgroup":35,"language":179,"tags":180,"starCount":182,"ownerUid":183,"latestTag":184,"platforms":185,"lastReleaseTime":186,"downloadCount":187,"description":188},"yetone","magpie",1383809012,"Go",[105,107,110,181,135,166],"gemini-cli",16805,1206493,"v0.1.1293",[87,88],"2026-10-19T12:31:24Z",309,"Magpie is a locally running AI gateway that lets coding agents such as Claude Code, Codex, Gemini CLI and OpenCode use models like Kimi, DeepSeek, GLM and ChatGPT, with one-click menu bar switching and automatic failover to the next account when quota runs out.",{"description":190,"readmeHtml":191},"Optimized vLLM Docker configuration and startup scripts for NVIDIA DGX Spark, supporting single-node and multi-node clusters using Ray or native PyTorch distributed mode, with InfiniBand\u002FRDMA (NCCL), high-performance model loading, and custom environment configuration.","\u003Ch2 id=\"project-overview\">Project Overview\u003C\u002Fh2>\u003Cp>This repository contains Docker configurations and startup scripts optimized for running vLLM on NVIDIA DGX Spark, from single-node to multi-node clusters using Ray or vLLM's native PyTorch distributed mode. It supports InfiniBand\u002FRDMA (NCCL), custom environment configuration, and high-performance model loading through fastsafetensors and InstantTensor. Cluster setups include direct connections between dual Sparks, QSFP\u002FRoCE switch configurations, and 3-node mesh configurations.\u003C\u002Fp>\n\u003Ch2 id=\"key-features\">Key Features\u003C\u002Fh2>\u003Cul>\n\u003Cli>\u003Cstrong>Multi-node inference\u003C\u002Fstrong>: Single-node and multi-node cluster deployment\u003C\u002Fli>\n\u003Cli>\u003Cstrong>High-performance model loading\u003C\u002Fstrong>: Optimized with fastsafetensors and InstantTensor\u003C\u002Fli>\n\u003Cli>\u003Cstrong>InfiniBand\u002FRDMA support\u003C\u002Fstrong>: High-speed interconnect via NCCL\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Multiple cluster configurations\u003C\u002Fstrong>: Dual Spark direct, QSFP\u002FRoCE switch, 3-node mesh\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Flexible build options\u003C\u002Fstrong>: Precompiled wheels, rebuilding vLLM or FlashInfer from source\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"target-audience\">Target Audience\u003C\u002Fh2>\u003Cul>\n\u003Cli>Developers using NVIDIA DGX Spark for LLM inference\u003C\u002Fli>\n\u003Cli>AI engineers needing multi-node distributed inference\u003C\u002Fli>\n\u003Cli>Researchers interested in high-performance vLLM deployments\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"quick-start\">Quick Start\u003C\u002Fh2>\u003Col>\n\u003Cli>Default image: \u003Ccode>eugr\u002Fspark-vllm:latest\u003C\u002Fcode>\u003C\u002Fli>\n\u003Cli>Run \u003Ccode>build-and-copy.sh\u003C\u002Fcode> to build and start the container\u003C\u002Fli>\n\u003Cli>For multi-node, configure passwordless SSH between Sparks\u003C\u002Fli>\n\u003Cli>Refer to the \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Feugr\u002Fspark-vllm-docker\u002Fblob\u002Fmain\u002Fdocs\u002FNETWORKING.md\" target=\"_blank\" rel=\"noopener noreferrer\">Networking Guide\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Ch2 id=\"core-advantages\">Core Advantages\u003C\u002Fh2>\u003Cul>\n\u003Cli>Optimized vLLM deployment for DGX Spark\u003C\u002Fli>\n\u003Cli>Multi-node distributed inference support\u003C\u002Fli>\n\u003Cli>Flexible build options for custom vLLM versions\u003C\u002Fli>\n\u003Cli>Community-driven open-source project (not affiliated with NVIDIA)\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"system-requirements\">System Requirements\u003C\u002Fh2>\u003Cul>\n\u003Cli>Hardware: NVIDIA DGX Spark\u003C\u002Fli>\n\u003Cli>Network: InfiniBand\u002FRDMA (optional), passwordless SSH (multi-node)\u003C\u002Fli>\n\u003C\u002Ful>\n",1792450905911]