[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f3m9wrm7vkyroj":3,"$fducyunnqimab":72,"$f39eq6ea7ma4g7":74,"$f3boz9rj14q7d5":324,"$f8ypa021er9qz":336,"$f1xoo4p4nkwctg":338},{"id":4,"name":5,"publisher":6,"paramsB":7,"useCase":8,"licenseStatus":9,"licenseUrl":10,"sourceUrl":11,"descriptionZh":12,"descriptionEn":13,"imageUrl":14,"downloadUrls":15,"stats":24,"variants":29,"derivations":30,"measuredHardware":33,"updatedAt":71},97,"NVIDIA-Nemotron-3.5-Bolt-E4B-4GB","NVIDIA",8,"CHAT","OPEN_WEIGHTS","https:\u002F\u002Fgithub.com\u002FOpenMDW\u002FOpenMDW\u002Fblob\u002Fmain\u002F1.1\u002FLICENSE.OpenMDW-1.1","https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FNVIDIA-Nemotron-3.5-Bolt-E4B-4GB","NVIDIA 于 2026-10-23 发布的 Nemotron 3.5 Bolt（Apache-2.0 的 Google Gemma-4-E4B-it 底座 + NVIDIA 后训练，官方后训练数据截止 2026-05）：8B 总参 \u002F 4B 有效参，面向本地语音与设备助手、交互人设、边缘嵌入式场景。主打多轮工具调用、指令遵循与输入接地（input grounding）；chat template 控制 thinking 模式与 conversational 模式，thinking 预算可调。官方 Q4_K_M GGUF 实测 **6.18GB**（仓库名里的「4GB」指 4B 有效参数量级），llama.cpp 官方给了 Docker 配方（ghcr.io\u002Fggml-org\u002Fllama.cpp:server-cuda13-b11459，-c 65536、--jinja），另配 DFlash 草稿模型做投机解码（--spec-type draft-dflash --spec-draft-n-max 7）。上下文输入\u002F输出各 64K token，支持 36 种语言。Q4_K_M 口径基准（对照 Gemma-4-E4B-it Q4_K_M \u002F Qwen3.5-4B Q4_K_M）：MultiChallenge 55.60、RoleBench 71.21、RoleMRC 79.12、ToolSandbox 74.13、IHEval 82.40、IFBench 64.00、StructEval-T 79.99、BFCL-v3 67.31、Tau2-Bench 52.32、AgentIF 22.95、RULER(64K) 85.76、LongMemEval 69.92、LongBench v2 (48K) 48.67——工具调用与角色扮演两项相对底座提升最明显。许可 OpenMDW-1.1（底座 Gemma 4 部分适用 Apache-2.0）。","NVIDIA's Nemotron 3.5 Bolt, released 2026-10-23 (Google Gemma-4-E4B-it on Apache-2.0, post-trained by NVIDIA, post-training data cutoff May 2026): 8B total parameters with 4B effective, aimed at local voice and device assistants, interactive personas and embedded edge interfaces. Its focus is multi-turn tool use, instruction following and input grounding; the chat template controls thinking mode and conversational mode, with a tunable thinking budget. The official Q4_K_M GGUF measures **6.18GB** (the \"4GB\" in the repo name refers to the 4B effective-parameter class), and NVIDIA ships a llama.cpp Docker recipe (ghcr.io\u002Fggml-org\u002Fllama.cpp:server-cuda13-b11459, -c 65536, --jinja) plus a DFlash draft model for speculative decoding (--spec-type draft-dflash --spec-draft-n-max 7). Context is 64K tokens in and out, across 36 languages. Q4_K_M benchmarks (against Gemma-4-E4B-it Q4_K_M \u002F Qwen3.5-4B Q4_K_M): MultiChallenge 55.60, RoleBench 71.21, RoleMRC 79.12, ToolSandbox 74.13, IHEval 82.40, IFBench 64.00, StructEval-T 79.99, BFCL-v3 67.31, Tau2-Bench 52.32, AgentIF 22.95, RULER (64K) 85.76, LongMemEval 69.92, LongBench v2 (48K) 48.67 — the clearest gains over the base model are in tool use and role play. License OpenMDW-1.1 (the Gemma 4 base remains under Apache-2.0).",null,[16,18,21],{"url":11,"label":17},"HuggingFace",{"url":19,"label":20},"https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FNVIDIA-Nemotron-3.5-Bolt-E4B-4GB-DFlash","DFlash Drafter",{"url":22,"label":23},"https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FNVIDIA-Nemotron-3.5-Bolt-E4B-BF16","BF16",{"recordCount":25,"variantCount":26,"hardwareCount":27,"unlinkedRecordCount":25,"evidence":28},26,0,14,{"L0":26,"L1":26,"L2":25},[],{"parents":31,"children":32},[],[],[34,38,41,43,45,48,51,53,55,57,60,63,65,69],{"name":35,"vramGb":36,"recordCount":37},"AMD Radeon RX 7900 XTX",24,2,{"name":39,"vramGb":40,"recordCount":37},"AMD Ryzen AI Max+ 395",128,{"name":42,"vramGb":14,"recordCount":37},"Apple M2 Ultra",{"name":44,"vramGb":40,"recordCount":37},"Apple M5 Max",{"name":46,"vramGb":47,"recordCount":37},"Apple M5 Ultra",512,{"name":49,"vramGb":50,"recordCount":37},"NVIDIA H100 80GB HBM3",80,{"name":52,"vramGb":4,"recordCount":37},"NVIDIA H20",{"name":54,"vramGb":36,"recordCount":37},"NVIDIA RTX 3090",{"name":56,"vramGb":36,"recordCount":37},"NVIDIA RTX 4090",{"name":58,"vramGb":59,"recordCount":37},"NVIDIA RTX 5090",32,{"name":61,"vramGb":62,"recordCount":37},"NVIDIA RTX PRO 6000 Blackwell 96GB",96,{"name":64,"vramGb":40,"recordCount":37},"NVIDIA RTX Spark",{"name":66,"vramGb":67,"recordCount":68},"NVIDIA RTX 4080 Super",16,1,{"name":70,"vramGb":67,"recordCount":68},"NVIDIA RTX 5080","2026-10-24T01:42:21.453Z",{"favorited":73},false,{"items":75,"total":25,"page":68,"pageSize":323},[76,93,106,117,127,137,148,159,167,176,185,194,202,213,221,231,239,248,257,266,275,284,291,300,307,316],{"modelId":4,"modelName":5,"hardwareId":68,"hardwareName":56,"frameworkId":68,"frameworkName":77,"quantization":78,"hardwareVramGb":36,"decodeTps":79,"prefillTps":80,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":82,"ttfbGb":83,"mtpAcceptanceRate":84,"powerW":85,"sampleCount":68,"reproduceCount":68,"sourceTypes":86,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":90,"latestOs":91,"latestCudaVersion":92,"latestFlashAttention":73,"latestOutputSpec":14},"llama.cpp","Q4_K_M",160.59,829.66,15.88,2.322,1.28,0.5392,324,[87],"OTHER","L2","2026-10-24T01:43:30.888Z","v0.36","Ubuntu 22.04","12.6",{"modelId":4,"modelName":5,"hardwareId":68,"hardwareName":56,"frameworkId":94,"frameworkName":95,"quantization":78,"hardwareVramGb":36,"decodeTps":96,"prefillTps":97,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":98,"ttfbGb":99,"mtpAcceptanceRate":100,"powerW":101,"sampleCount":68,"reproduceCount":68,"sourceTypes":102,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":103,"latestOs":104,"latestCudaVersion":92,"latestFlashAttention":105,"latestOutputSpec":14},9,"Strata",225.92,741.96,1.956,1.81,0.4885,304.8,[87],"v0.44","Windows 11 24H2",true,{"modelId":4,"modelName":5,"hardwareId":37,"hardwareName":54,"frameworkId":68,"frameworkName":77,"quantization":78,"hardwareVramGb":36,"decodeTps":107,"prefillTps":108,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":109,"ttfbGb":110,"mtpAcceptanceRate":111,"powerW":112,"sampleCount":68,"reproduceCount":68,"sourceTypes":113,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":114,"latestOs":115,"latestCudaVersion":116,"latestFlashAttention":73,"latestOutputSpec":14},94.58,353.95,0.997,0.76,0.7312,286.1,[87],"v0.30","Windows 11 23H2","12.8",{"modelId":4,"modelName":5,"hardwareId":37,"hardwareName":54,"frameworkId":94,"frameworkName":95,"quantization":78,"hardwareVramGb":36,"decodeTps":118,"prefillTps":119,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":120,"ttfbGb":121,"mtpAcceptanceRate":122,"powerW":123,"sampleCount":68,"reproduceCount":68,"sourceTypes":124,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":125,"latestOs":91,"latestCudaVersion":126,"latestFlashAttention":73,"latestOutputSpec":14},125.83,630.14,2.193,1.01,0.5811,250.4,[87],"v0.28","12.4",{"modelId":4,"modelName":5,"hardwareId":128,"hardwareName":35,"frameworkId":68,"frameworkName":77,"quantization":78,"hardwareVramGb":36,"decodeTps":129,"prefillTps":130,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":131,"ttfbGb":132,"mtpAcceptanceRate":133,"powerW":134,"sampleCount":68,"reproduceCount":68,"sourceTypes":135,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":136,"latestOs":115,"latestCudaVersion":126,"latestFlashAttention":105,"latestOutputSpec":14},11,102.56,353.32,2.421,0.82,0.6166,210.1,[87],"v0.35",{"modelId":4,"modelName":5,"hardwareId":128,"hardwareName":35,"frameworkId":138,"frameworkName":139,"quantization":78,"hardwareVramGb":36,"decodeTps":140,"prefillTps":141,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":142,"ttfbGb":143,"mtpAcceptanceRate":14,"powerW":144,"sampleCount":68,"reproduceCount":68,"sourceTypes":145,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":146,"latestOs":147,"latestCudaVersion":92,"latestFlashAttention":105,"latestOutputSpec":14},3,"Ollama",86.87,421.48,1.935,0.69,217.1,[87],"v0.24","Ubuntu 24.04",{"modelId":4,"modelName":5,"hardwareId":149,"hardwareName":58,"frameworkId":37,"frameworkName":150,"quantization":23,"hardwareVramGb":59,"decodeTps":151,"prefillTps":152,"vramGb":153,"genSeconds":14,"modelUseCase":8,"ttftS":154,"ttfbGb":155,"mtpAcceptanceRate":14,"powerW":156,"sampleCount":68,"reproduceCount":68,"sourceTypes":157,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":158,"latestOs":91,"latestCudaVersion":116,"latestFlashAttention":105,"latestOutputSpec":14},13,"vLLM",171.53,804.96,30.4,0.248,1.37,435.1,[87],"v0.50",{"modelId":4,"modelName":5,"hardwareId":149,"hardwareName":58,"frameworkId":138,"frameworkName":139,"quantization":78,"hardwareVramGb":59,"decodeTps":160,"prefillTps":161,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":162,"ttfbGb":163,"mtpAcceptanceRate":14,"powerW":164,"sampleCount":68,"reproduceCount":68,"sourceTypes":165,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":166,"latestOs":115,"latestCudaVersion":92,"latestFlashAttention":105,"latestOutputSpec":14},279.75,1289.8,1.235,2.24,466.6,[87],"v0.32",{"modelId":4,"modelName":5,"hardwareId":27,"hardwareName":70,"frameworkId":138,"frameworkName":139,"quantization":78,"hardwareVramGb":67,"decodeTps":168,"prefillTps":169,"vramGb":170,"genSeconds":14,"modelUseCase":8,"ttftS":171,"ttfbGb":172,"mtpAcceptanceRate":14,"powerW":173,"sampleCount":68,"reproduceCount":68,"sourceTypes":174,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":175,"latestOs":147,"latestCudaVersion":116,"latestFlashAttention":105,"latestOutputSpec":14},181.19,808.04,15.2,1.575,1.45,281,[87],"v0.47",{"modelId":4,"modelName":5,"hardwareId":177,"hardwareName":66,"frameworkId":68,"frameworkName":77,"quantization":78,"hardwareVramGb":67,"decodeTps":178,"prefillTps":179,"vramGb":170,"genSeconds":14,"modelUseCase":8,"ttftS":180,"ttfbGb":181,"mtpAcceptanceRate":182,"powerW":183,"sampleCount":68,"reproduceCount":68,"sourceTypes":184,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":125,"latestOs":104,"latestCudaVersion":116,"latestFlashAttention":73,"latestOutputSpec":14},20,162.68,521.64,1.222,1.3,0.635,274.3,[87],{"modelId":4,"modelName":5,"hardwareId":36,"hardwareName":39,"frameworkId":68,"frameworkName":77,"quantization":78,"hardwareVramGb":40,"decodeTps":186,"prefillTps":187,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":188,"ttfbGb":189,"mtpAcceptanceRate":190,"powerW":191,"sampleCount":68,"reproduceCount":68,"sourceTypes":192,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":193,"latestOs":115,"latestCudaVersion":92,"latestFlashAttention":105,"latestOutputSpec":14},35.76,167.23,0.588,0.29,0.6893,45.3,[87],"v0.31",{"modelId":4,"modelName":5,"hardwareId":36,"hardwareName":39,"frameworkId":138,"frameworkName":139,"quantization":78,"hardwareVramGb":40,"decodeTps":195,"prefillTps":196,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":197,"ttfbGb":198,"mtpAcceptanceRate":14,"powerW":199,"sampleCount":68,"reproduceCount":68,"sourceTypes":200,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":201,"latestOs":147,"latestCudaVersion":126,"latestFlashAttention":105,"latestOutputSpec":14},33.77,108.98,1.791,0.27,64.3,[87],"v0.20",{"modelId":4,"modelName":5,"hardwareId":203,"hardwareName":44,"frameworkId":68,"frameworkName":77,"quantization":78,"hardwareVramGb":40,"decodeTps":204,"prefillTps":205,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":206,"ttfbGb":207,"mtpAcceptanceRate":208,"powerW":209,"sampleCount":68,"reproduceCount":68,"sourceTypes":210,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":211,"latestOs":212,"latestCudaVersion":14,"latestFlashAttention":73,"latestOutputSpec":14},28,76.62,265.83,1.738,0.61,0.6602,46.4,[87],"v0.25","macOS 26.1",{"modelId":4,"modelName":5,"hardwareId":203,"hardwareName":44,"frameworkId":138,"frameworkName":139,"quantization":78,"hardwareVramGb":40,"decodeTps":214,"prefillTps":215,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":216,"ttfbGb":217,"mtpAcceptanceRate":14,"powerW":218,"sampleCount":68,"reproduceCount":68,"sourceTypes":219,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":220,"latestOs":212,"latestCudaVersion":14,"latestFlashAttention":105,"latestOutputSpec":14},69.91,321.78,1.461,0.56,48.4,[87],"v0.21",{"modelId":4,"modelName":5,"hardwareId":222,"hardwareName":46,"frameworkId":68,"frameworkName":77,"quantization":78,"hardwareVramGb":47,"decodeTps":223,"prefillTps":224,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":225,"ttfbGb":226,"mtpAcceptanceRate":227,"powerW":228,"sampleCount":68,"reproduceCount":68,"sourceTypes":229,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":230,"latestOs":212,"latestCudaVersion":14,"latestFlashAttention":105,"latestOutputSpec":14},29,130.96,547.83,1.483,1.05,0.6362,135.6,[87],"v0.45",{"modelId":4,"modelName":5,"hardwareId":222,"hardwareName":46,"frameworkId":138,"frameworkName":139,"quantization":78,"hardwareVramGb":47,"decodeTps":232,"prefillTps":233,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":234,"ttfbGb":235,"mtpAcceptanceRate":14,"powerW":236,"sampleCount":68,"reproduceCount":68,"sourceTypes":237,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":238,"latestOs":212,"latestCudaVersion":14,"latestFlashAttention":105,"latestOutputSpec":14},106.35,353.45,0.543,0.85,102.9,[87],"v0.43",{"modelId":4,"modelName":5,"hardwareId":240,"hardwareName":49,"frameworkId":138,"frameworkName":139,"quantization":78,"hardwareVramGb":50,"decodeTps":241,"prefillTps":242,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":243,"ttfbGb":244,"mtpAcceptanceRate":14,"powerW":245,"sampleCount":68,"reproduceCount":68,"sourceTypes":246,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":230,"latestOs":91,"latestCudaVersion":247,"latestFlashAttention":73,"latestOutputSpec":14},55,345.26,1726.58,1.805,2.76,678.6,[87],"13.0",{"modelId":4,"modelName":5,"hardwareId":240,"hardwareName":49,"frameworkId":94,"frameworkName":95,"quantization":78,"hardwareVramGb":50,"decodeTps":249,"prefillTps":250,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":251,"ttfbGb":252,"mtpAcceptanceRate":253,"powerW":254,"sampleCount":68,"reproduceCount":68,"sourceTypes":255,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":256,"latestOs":91,"latestCudaVersion":116,"latestFlashAttention":105,"latestOutputSpec":14},317.23,1729.64,1.217,2.54,0.5148,649.6,[87],"v0.41",{"modelId":4,"modelName":5,"hardwareId":258,"hardwareName":52,"frameworkId":68,"frameworkName":77,"quantization":78,"hardwareVramGb":4,"decodeTps":259,"prefillTps":260,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":261,"ttfbGb":262,"mtpAcceptanceRate":263,"powerW":264,"sampleCount":68,"reproduceCount":68,"sourceTypes":265,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":103,"latestOs":91,"latestCudaVersion":126,"latestFlashAttention":105,"latestOutputSpec":14},56,61.75,319.38,1.567,0.49,0.5515,366.4,[87],{"modelId":4,"modelName":5,"hardwareId":258,"hardwareName":52,"frameworkId":94,"frameworkName":95,"quantization":78,"hardwareVramGb":4,"decodeTps":267,"prefillTps":268,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":269,"ttfbGb":270,"mtpAcceptanceRate":271,"powerW":272,"sampleCount":68,"reproduceCount":68,"sourceTypes":273,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":274,"latestOs":104,"latestCudaVersion":116,"latestFlashAttention":105,"latestOutputSpec":14},91.09,326.32,0.101,0.73,0.5315,345.4,[87],"v0.40",{"modelId":4,"modelName":5,"hardwareId":276,"hardwareName":61,"frameworkId":37,"frameworkName":150,"quantization":23,"hardwareVramGb":62,"decodeTps":277,"prefillTps":278,"vramGb":279,"genSeconds":14,"modelUseCase":8,"ttftS":280,"ttfbGb":281,"mtpAcceptanceRate":14,"powerW":282,"sampleCount":68,"reproduceCount":68,"sourceTypes":283,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":90,"latestOs":115,"latestCudaVersion":116,"latestFlashAttention":105,"latestOutputSpec":14},68,177.62,884.45,31.5,0.72,1.42,350.6,[87],{"modelId":4,"modelName":5,"hardwareId":276,"hardwareName":61,"frameworkId":138,"frameworkName":139,"quantization":78,"hardwareVramGb":62,"decodeTps":285,"prefillTps":286,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":287,"ttfbGb":288,"mtpAcceptanceRate":14,"powerW":289,"sampleCount":68,"reproduceCount":68,"sourceTypes":290,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":90,"latestOs":147,"latestCudaVersion":92,"latestFlashAttention":105,"latestOutputSpec":14},264.78,915.82,0.581,2.12,391.8,[87],{"modelId":4,"modelName":5,"hardwareId":292,"hardwareName":42,"frameworkId":68,"frameworkName":77,"quantization":78,"hardwareVramGb":14,"decodeTps":293,"prefillTps":294,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":295,"ttfbGb":296,"mtpAcceptanceRate":297,"powerW":298,"sampleCount":68,"reproduceCount":68,"sourceTypes":299,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":114,"latestOs":212,"latestCudaVersion":14,"latestFlashAttention":105,"latestOutputSpec":14},103,119.47,397.94,2.117,0.96,0.6578,135.9,[87],{"modelId":4,"modelName":5,"hardwareId":292,"hardwareName":42,"frameworkId":138,"frameworkName":139,"quantization":78,"hardwareVramGb":14,"decodeTps":301,"prefillTps":302,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":303,"ttfbGb":304,"mtpAcceptanceRate":14,"powerW":305,"sampleCount":68,"reproduceCount":68,"sourceTypes":306,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":114,"latestOs":212,"latestCudaVersion":14,"latestFlashAttention":105,"latestOutputSpec":14},118.08,636.62,1.393,0.94,126.1,[87],{"modelId":4,"modelName":5,"hardwareId":308,"hardwareName":64,"frameworkId":68,"frameworkName":77,"quantization":78,"hardwareVramGb":40,"decodeTps":309,"prefillTps":310,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":311,"ttfbGb":312,"mtpAcceptanceRate":313,"powerW":314,"sampleCount":68,"reproduceCount":68,"sourceTypes":315,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":238,"latestOs":147,"latestCudaVersion":247,"latestFlashAttention":73,"latestOutputSpec":14},110,80.69,270.65,0.226,0.65,0.5739,55.8,[87],{"modelId":4,"modelName":5,"hardwareId":308,"hardwareName":64,"frameworkId":138,"frameworkName":139,"quantization":78,"hardwareVramGb":40,"decodeTps":317,"prefillTps":318,"vramGb":81,"genSeconds":14,"modelUseCase":8,"ttftS":319,"ttfbGb":320,"mtpAcceptanceRate":14,"powerW":321,"sampleCount":68,"reproduceCount":68,"sourceTypes":322,"evidenceLevel":88,"updatedAt":89,"latestFrameworkVersion":114,"latestOs":147,"latestCudaVersion":247,"latestFlashAttention":73,"latestOutputSpec":14},79.19,309.44,0.666,0.63,65.4,[87],50,{"items":325,"total":68,"page":68,"pageSize":177},[326],{"id":327,"type":328,"title":329,"excerpt":330,"author":331,"replyCount":138,"createdAt":333,"lastRepliedAt":14,"resolvedAt":14,"editedAt":14,"model":334,"hardware":335,"variant":14,"record":14},44,"HELP","Nemotron 3.5 Bolt-E4B（6.18GB Q4_K_M）做本地语音\u002F设备助手，DFlash 值不值得开？","NVIDIA 同一天发的 Nemotron 3.5 里，Bolt-E4B 是我真正会下手的那个：8B 总参 \u002F 4B 有效参，Q4KM GGUF 官方实测 6.18GB，64K 上下文、36 种语言，定位就是本地语音与设备助手。站内昨天已经",{"nickname":332,"avatarUrl":14},"江南卡皇","2026-10-24T01:48:47.085Z",{"name":5},{"name":58},{"items":337},[],{"items":339,"total":68,"page":68,"pageSize":138},[340],{"slug":341,"category":342,"publishedAt":343,"titleZh":344,"titleEn":345,"summaryZh":346,"summaryEn":347,"models":348,"hardwares":355},"nvidia-nemotron-35-super-vl-bolt-2026-10-24","NEWS","2026-10-24T14:00:00.000Z","NVIDIA 同日双发 Nemotron 3.5：120B-A12B 的 Super VL 与 6GB 边缘模型 Bolt","NVIDIA ships two Nemotron 3.5 models in one day: the 120B-A12B Super VL and a 6GB edge model, Bolt","10 月 23 日 NVIDIA 一次放出两个 Nemotron 3.5 模型。Super-VL-120B-A12B-NVFP4：Mamba-2 与 Transformer 混合 MoE + C-RADIO 视觉编码器，1M 上下文，NVFP4 权重实测 75.0GB，官方部署到 DGX Spark \u002F RTX PRO 6000 \u002F H200 \u002F GB200。Bolt-E4B-4GB：8B-A4B 的边缘小模型，Q4_K_M GGUF 仅 6.18GB，主打工具调用与设备侧语音助手。站内新增 103 条实测。","NVIDIA released two Nemotron 3.5 models on the same day, Oct 23. Super-VL-120B-A12B-NVFP4 mixes Mamba-2 and Transformer layers in a hybrid MoE with a C-RADIO vision encoder, carries a 1M context and measures 75.0GB of NVFP4 weights, and deploys officially on DGX Spark \u002F RTX PRO 6000 \u002F H200 \u002F GB200. Bolt-E4B-4GB is the 8B-A4B edge sibling, a 6.18GB Q4_K_M GGUF aimed at tool use and on-device voice assistants. The batch adds 103 new runs on-site.",[349,352,354],{"id":350,"name":351},67,"gemma-4-E4B-it",{"id":62,"name":353},"NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4",{"id":4,"name":5},[356,357,360],{"id":149,"name":58},{"id":358,"name":359},54,"NVIDIA DGX Spark",{"id":276,"name":61}]