[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f3vbtd029tzi2y":3,"$f22jsjvoej42y8":62,"$f3cetsn5rtn1c6":64,"$fxtn8y166jicm":279,"$f2b92ed1sb7b8f":295,"$f8ypa021er9qz":325},{"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":21,"variants":26,"derivations":27,"measuredHardware":30,"updatedAt":61},7,"DeepSeek-V4-Flash","DeepSeek",284,"CHAT","OPEN_SOURCE","https:\u002F\u002Fhuggingface.co\u002Fdeepseek-ai\u002FDeepSeek-V4-Flash\u002Fblob\u002Fmain\u002FLICENSE","https:\u002F\u002Fhuggingface.co\u002Fdeepseek-ai\u002FDeepSeek-V4-Flash","DeepSeek-V4 系列的 MoE 模型：284B 总参数 \u002F 13B 激活，1M 上下文，MIT 许可。混合注意力（CSA + HCA）把长上下文的 KV cache 压到 V3.2 的约十分之一。","A DeepSeek-V4 MoE model: 284B total parameters with 13B activated, 1M context, MIT licensed. Its hybrid attention (CSA + HCA) cuts long-context KV cache to roughly a tenth of V3.2's.","\u002Fuploads\u002Fmodels\u002Fm7-ff514545cead.png",[16,18],{"url":11,"label":17},"HuggingFace",{"url":19,"label":20},"https:\u002F\u002Fmodelscope.cn\u002Fmodels\u002Fdeepseek-ai\u002FDeepSeek-V4-Flash","ModelScope",{"recordCount":22,"variantCount":23,"hardwareCount":24,"unlinkedRecordCount":22,"evidence":25},32,0,11,{"L0":22,"L1":23,"L2":23},[],{"parents":28,"children":29},[],[],[31,34,37,40,42,46,49,52,54,57,59],{"name":32,"vramGb":33,"recordCount":4},"NVIDIA RTX 5070 Ti",16,{"name":35,"vramGb":33,"recordCount":36},"NVIDIA RTX 4080 Super",6,{"name":38,"vramGb":39,"recordCount":36},"NVIDIA RTX 4090",24,{"name":41,"vramGb":22,"recordCount":36},"NVIDIA RTX 5090",{"name":43,"vramGb":44,"recordCount":45},"2x NVIDIA RTX 5090",64,1,{"name":47,"vramGb":48,"recordCount":45},"4x NVIDIA RTX 3090",96,{"name":50,"vramGb":51,"recordCount":45},"Apple M4 Max",128,{"name":53,"vramGb":51,"recordCount":45},"Apple M5 Max",{"name":55,"vramGb":56,"recordCount":45},"NVIDIA RTX 3060 12GB",12,{"name":58,"vramGb":39,"recordCount":45},"NVIDIA RTX 3090",{"name":60,"vramGb":56,"recordCount":45},"NVIDIA RTX 5070","2026-09-24T12:55:08.665Z",{"favorited":63},false,{"items":65,"total":22,"page":45,"pageSize":278},[66,83,90,97,105,111,115,122,128,131,134,140,146,153,159,165,171,177,180,186,189,195,201,209,213,221,227,237,244,250,256,268],{"modelId":4,"modelName":5,"hardwareId":67,"hardwareName":35,"frameworkId":68,"frameworkName":69,"quantization":70,"hardwareVramGb":33,"decodeTps":71,"prefillTps":72,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":74,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":75,"evidenceLevel":77,"updatedAt":78,"latestFrameworkVersion":79,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},20,2,"vLLM","Q4_K_M",3.6,27,null,3.7,[76],"OTHER","L0","2026-10-06T12:06:15.223Z","0.7.3","Linux","12.6",true,{"modelId":4,"modelName":5,"hardwareId":67,"hardwareName":35,"frameworkId":68,"frameworkName":69,"quantization":84,"hardwareVramGb":33,"decodeTps":85,"prefillTps":86,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":87,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":88,"evidenceLevel":77,"updatedAt":89,"latestFrameworkVersion":79,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},"IQ3_S",5.4,40.5,2.47,[76],"2026-10-06T12:06:15.215Z",{"modelId":4,"modelName":5,"hardwareId":67,"hardwareName":35,"frameworkId":68,"frameworkName":69,"quantization":91,"hardwareVramGb":33,"decodeTps":92,"prefillTps":93,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":94,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":95,"evidenceLevel":77,"updatedAt":96,"latestFrameworkVersion":79,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},"IQ4_XS",4.1,30.7,3.26,[76],"2026-10-06T12:06:15.208Z",{"modelId":4,"modelName":5,"hardwareId":67,"hardwareName":35,"frameworkId":45,"frameworkName":98,"quantization":70,"hardwareVramGb":33,"decodeTps":99,"prefillTps":100,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":101,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":102,"evidenceLevel":77,"updatedAt":103,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},"llama.cpp",3.3,24.8,4.03,[76],"2026-10-06T12:06:15.200Z","b4200",{"modelId":4,"modelName":5,"hardwareId":67,"hardwareName":35,"frameworkId":45,"frameworkName":98,"quantization":84,"hardwareVramGb":33,"decodeTps":106,"prefillTps":107,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":108,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":109,"evidenceLevel":77,"updatedAt":110,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},5,37.5,2.67,[76],"2026-10-06T12:06:15.191Z",{"modelId":4,"modelName":5,"hardwareId":67,"hardwareName":35,"frameworkId":45,"frameworkName":98,"quantization":91,"hardwareVramGb":33,"decodeTps":74,"prefillTps":112,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":71,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":113,"evidenceLevel":77,"updatedAt":114,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},27.8,[76],"2026-10-06T12:06:15.183Z",{"modelId":4,"modelName":5,"hardwareId":116,"hardwareName":32,"frameworkId":68,"frameworkName":69,"quantization":70,"hardwareVramGb":33,"decodeTps":117,"prefillTps":118,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":119,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":120,"evidenceLevel":77,"updatedAt":121,"latestFrameworkVersion":79,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},17,4.4,33,3.03,[76],"2026-10-06T12:06:15.175Z",{"modelId":4,"modelName":5,"hardwareId":116,"hardwareName":32,"frameworkId":68,"frameworkName":69,"quantization":84,"hardwareVramGb":33,"decodeTps":123,"prefillTps":124,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":125,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":126,"evidenceLevel":77,"updatedAt":127,"latestFrameworkVersion":79,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},6.6,49.5,2.02,[76],"2026-10-06T12:06:15.165Z",{"modelId":4,"modelName":5,"hardwareId":116,"hardwareName":32,"frameworkId":68,"frameworkName":69,"quantization":91,"hardwareVramGb":33,"decodeTps":106,"prefillTps":107,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":108,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":129,"evidenceLevel":77,"updatedAt":130,"latestFrameworkVersion":79,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},[76],"2026-10-06T12:06:15.155Z",{"modelId":4,"modelName":5,"hardwareId":116,"hardwareName":32,"frameworkId":45,"frameworkName":98,"quantization":70,"hardwareVramGb":33,"decodeTps":92,"prefillTps":93,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":94,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":132,"evidenceLevel":77,"updatedAt":133,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},[76],"2026-10-06T12:06:15.145Z",{"modelId":4,"modelName":5,"hardwareId":116,"hardwareName":32,"frameworkId":45,"frameworkName":98,"quantization":84,"hardwareVramGb":33,"decodeTps":135,"prefillTps":136,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":137,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":138,"evidenceLevel":77,"updatedAt":139,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},6.1,45.8,2.18,[76],"2026-10-06T12:06:15.138Z",{"modelId":4,"modelName":5,"hardwareId":116,"hardwareName":32,"frameworkId":45,"frameworkName":98,"quantization":91,"hardwareVramGb":33,"decodeTps":141,"prefillTps":142,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":143,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":144,"evidenceLevel":77,"updatedAt":145,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},4.6,34.5,2.9,[76],"2026-10-06T12:06:15.130Z",{"modelId":4,"modelName":5,"hardwareId":147,"hardwareName":41,"frameworkId":68,"frameworkName":69,"quantization":70,"hardwareVramGb":22,"decodeTps":148,"prefillTps":149,"vramGb":22,"genSeconds":73,"modelUseCase":8,"ttftS":150,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":151,"evidenceLevel":77,"updatedAt":152,"latestFrameworkVersion":79,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},13,8.8,66,1.52,[76],"2026-10-06T12:06:15.123Z",{"modelId":4,"modelName":5,"hardwareId":147,"hardwareName":41,"frameworkId":68,"frameworkName":69,"quantization":91,"hardwareVramGb":22,"decodeTps":154,"prefillTps":155,"vramGb":22,"genSeconds":73,"modelUseCase":8,"ttftS":156,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":157,"evidenceLevel":77,"updatedAt":158,"latestFrameworkVersion":79,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},9.9,74.3,1.35,[76],"2026-10-06T12:06:15.110Z",{"modelId":4,"modelName":5,"hardwareId":147,"hardwareName":41,"frameworkId":45,"frameworkName":98,"quantization":70,"hardwareVramGb":22,"decodeTps":160,"prefillTps":161,"vramGb":22,"genSeconds":73,"modelUseCase":8,"ttftS":162,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":163,"evidenceLevel":77,"updatedAt":164,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},8.1,60.8,1.64,[76],"2026-10-06T12:06:15.102Z",{"modelId":4,"modelName":5,"hardwareId":147,"hardwareName":41,"frameworkId":45,"frameworkName":98,"quantization":84,"hardwareVramGb":22,"decodeTps":166,"prefillTps":167,"vramGb":22,"genSeconds":73,"modelUseCase":8,"ttftS":168,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":169,"evidenceLevel":77,"updatedAt":170,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},12.2,91.5,1.09,[76],"2026-10-06T12:06:15.094Z",{"modelId":4,"modelName":5,"hardwareId":147,"hardwareName":41,"frameworkId":45,"frameworkName":98,"quantization":91,"hardwareVramGb":22,"decodeTps":172,"prefillTps":173,"vramGb":22,"genSeconds":73,"modelUseCase":8,"ttftS":174,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":175,"evidenceLevel":77,"updatedAt":176,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},9.1,68.3,1.46,[76],"2026-10-06T12:06:15.084Z",{"modelId":4,"modelName":5,"hardwareId":45,"hardwareName":38,"frameworkId":68,"frameworkName":69,"quantization":70,"hardwareVramGb":39,"decodeTps":106,"prefillTps":107,"vramGb":39,"genSeconds":73,"modelUseCase":8,"ttftS":108,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":178,"evidenceLevel":77,"updatedAt":179,"latestFrameworkVersion":79,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},[76],"2026-10-06T12:06:15.075Z",{"modelId":4,"modelName":5,"hardwareId":45,"hardwareName":38,"frameworkId":68,"frameworkName":69,"quantization":91,"hardwareVramGb":39,"decodeTps":181,"prefillTps":182,"vramGb":39,"genSeconds":73,"modelUseCase":8,"ttftS":183,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":184,"evidenceLevel":77,"updatedAt":185,"latestFrameworkVersion":79,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},5.6,42,2.38,[76],"2026-10-06T12:06:15.064Z",{"modelId":4,"modelName":5,"hardwareId":45,"hardwareName":38,"frameworkId":45,"frameworkName":98,"quantization":70,"hardwareVramGb":39,"decodeTps":141,"prefillTps":142,"vramGb":39,"genSeconds":73,"modelUseCase":8,"ttftS":143,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":187,"evidenceLevel":77,"updatedAt":188,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},[76],"2026-10-06T12:06:15.056Z",{"modelId":4,"modelName":5,"hardwareId":45,"hardwareName":38,"frameworkId":45,"frameworkName":98,"quantization":84,"hardwareVramGb":39,"decodeTps":190,"prefillTps":191,"vramGb":39,"genSeconds":73,"modelUseCase":8,"ttftS":192,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":193,"evidenceLevel":77,"updatedAt":194,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},6.8,51,1.96,[76],"2026-10-06T12:06:15.046Z",{"modelId":4,"modelName":5,"hardwareId":45,"hardwareName":38,"frameworkId":45,"frameworkName":98,"quantization":91,"hardwareVramGb":39,"decodeTps":196,"prefillTps":197,"vramGb":39,"genSeconds":73,"modelUseCase":8,"ttftS":198,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":199,"evidenceLevel":77,"updatedAt":200,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},5.1,38.3,2.61,[76],"2026-10-06T12:06:15.033Z",{"modelId":4,"modelName":5,"hardwareId":202,"hardwareName":53,"frameworkId":45,"frameworkName":98,"quantization":84,"hardwareVramGb":51,"decodeTps":203,"prefillTps":204,"vramGb":205,"genSeconds":73,"modelUseCase":8,"ttftS":206,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":207,"evidenceLevel":77,"updatedAt":208,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},28,4.2,31.5,115.5,3.17,[76],"2026-10-06T03:39:18.254Z",{"modelId":4,"modelName":5,"hardwareId":210,"hardwareName":50,"frameworkId":45,"frameworkName":98,"quantization":84,"hardwareVramGb":51,"decodeTps":74,"prefillTps":112,"vramGb":205,"genSeconds":73,"modelUseCase":8,"ttftS":71,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":211,"evidenceLevel":77,"updatedAt":212,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},26,[76],"2026-10-06T03:39:18.245Z",{"modelId":4,"modelName":5,"hardwareId":214,"hardwareName":55,"frameworkId":45,"frameworkName":98,"quantization":215,"hardwareVramGb":56,"decodeTps":216,"prefillTps":217,"vramGb":56,"genSeconds":73,"modelUseCase":8,"ttftS":218,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":219,"evidenceLevel":77,"updatedAt":220,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},15,"IQ2_XXS",3.5,26.3,3.8,[76],"2026-10-06T03:39:17.893Z",{"modelId":4,"modelName":5,"hardwareId":68,"hardwareName":58,"frameworkId":45,"frameworkName":98,"quantization":84,"hardwareVramGb":39,"decodeTps":222,"prefillTps":223,"vramGb":39,"genSeconds":73,"modelUseCase":8,"ttftS":224,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":225,"evidenceLevel":77,"updatedAt":226,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},6.3,47.3,2.11,[76],"2026-10-06T03:39:17.885Z",{"modelId":4,"modelName":5,"hardwareId":116,"hardwareName":32,"frameworkId":228,"frameworkName":229,"quantization":230,"hardwareVramGb":33,"decodeTps":231,"prefillTps":232,"vramGb":33,"genSeconds":73,"modelUseCase":8,"ttftS":233,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":234,"evidenceLevel":77,"updatedAt":235,"latestFrameworkVersion":236,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},9,"Strata","IQ2_XS",9.3,69.8,1.43,[76],"2026-10-06T03:39:17.877Z","0.5.2",{"modelId":4,"modelName":5,"hardwareId":238,"hardwareName":60,"frameworkId":45,"frameworkName":98,"quantization":230,"hardwareVramGb":56,"decodeTps":239,"prefillTps":240,"vramGb":56,"genSeconds":73,"modelUseCase":8,"ttftS":241,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":242,"evidenceLevel":77,"updatedAt":243,"latestFrameworkVersion":104,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},18,5.5,41.3,2.42,[76],"2026-10-06T03:39:17.867Z",{"modelId":4,"modelName":5,"hardwareId":147,"hardwareName":41,"frameworkId":68,"frameworkName":69,"quantization":84,"hardwareVramGb":22,"decodeTps":245,"prefillTps":246,"vramGb":22,"genSeconds":73,"modelUseCase":8,"ttftS":247,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":248,"evidenceLevel":77,"updatedAt":249,"latestFrameworkVersion":79,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},13.2,99,1.01,[76],"2026-10-06T03:39:17.860Z",{"modelId":4,"modelName":5,"hardwareId":45,"hardwareName":38,"frameworkId":68,"frameworkName":69,"quantization":84,"hardwareVramGb":39,"decodeTps":251,"prefillTps":252,"vramGb":39,"genSeconds":73,"modelUseCase":8,"ttftS":253,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":254,"evidenceLevel":77,"updatedAt":255,"latestFrameworkVersion":79,"latestOs":80,"latestCudaVersion":81,"latestFlashAttention":82,"latestOutputSpec":73},7.5,56.3,1.78,[76],"2026-10-06T03:39:17.851Z",{"modelId":4,"modelName":5,"hardwareId":36,"hardwareName":43,"frameworkId":45,"frameworkName":98,"quantization":257,"hardwareVramGb":44,"decodeTps":258,"prefillTps":259,"vramGb":260,"genSeconds":73,"modelUseCase":8,"ttftS":261,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":262,"evidenceLevel":77,"updatedAt":264,"latestFrameworkVersion":265,"latestOs":266,"latestCudaVersion":267,"latestFlashAttention":73,"latestOutputSpec":73},"UD-IQ2_XXS",19.39,692.41,51.17,0.117,[263],"GITHUB","2026-09-27T11:51:40.829Z","b10236","Ubuntu 24.04.4 LTS","13.3",{"modelId":4,"modelName":5,"hardwareId":269,"hardwareName":47,"frameworkId":45,"frameworkName":98,"quantization":257,"hardwareVramGb":48,"decodeTps":270,"prefillTps":271,"vramGb":272,"genSeconds":73,"modelUseCase":8,"ttftS":273,"ttfbGb":73,"mtpAcceptanceRate":73,"powerW":73,"sampleCount":45,"reproduceCount":45,"sourceTypes":274,"evidenceLevel":77,"updatedAt":275,"latestFrameworkVersion":276,"latestOs":266,"latestCudaVersion":277,"latestFlashAttention":73,"latestOutputSpec":73},10,35.68,420.73,87.16,0.111,[263],"2026-09-27T11:51:40.825Z","club3090-moecache-e0a2089","13.2",50,{"items":280,"total":45,"page":45,"pageSize":67},[281],{"id":282,"type":283,"title":284,"excerpt":285,"author":286,"replyCount":282,"createdAt":288,"lastRepliedAt":289,"resolvedAt":73,"editedAt":73,"model":290,"hardware":291,"variant":73,"record":292},3,"GUIDE","DeepSeek-V4-Flash 部署笔记：2×5090 与 4×3090 上 UD-IQ2_XXS 的对比","最近在两台机器上都部署了 DeepSeek-V4-Flash 的 UD-IQ2_XXS（这模型 FP8 原版太大，IQ2 档是甜点位），站内两条记录都是我提交的，这里展开讲讲。 2×5090（站内记录 #21）： decode 19.39 ",{"nickname":287,"avatarUrl":73},"bit_squeezer","2026-09-25T06:40:00.000Z","2026-09-27T00:33:00.000Z",{"name":5},{"name":43},{"id":293,"status":294,"published":82,"modelName":5,"hardwareName":43,"frameworkName":98,"quantization":257},21,"PUBLISHED",{"items":296,"total":45,"page":45,"pageSize":282},[297],{"slug":298,"category":299,"publishedAt":300,"titleZh":301,"titleEn":302,"summaryZh":303,"summaryEn":304,"models":305,"hardwares":323},"open-source-llm-5way-2026-10-09","NEWS","2026-10-09T16:00:00.000Z","开源大模型 5 家横评：Qwen 3.x \u002F DeepSeek V4.1 \u002F GLM-5.3 \u002F Llama 4.5 \u002F Gemma 4","Open-source LLM 5-way comparison: Qwen 3.x \u002F DeepSeek V4.1 \u002F GLM-5.3 \u002F Llama 4.5 \u002F Gemma 4","5 家开源旗舰横评，基于本地 DB RTX 4090 实测：Qwen 3.5 27B 91 tok\u002Fs、Gemma 4 12B 89 tok\u002Fs、Qwen 3.8 Flash 180B 12 tok\u002Fs、DeepSeek V4.1 763B 3 tok\u002Fs。27B 是单卡甜蜜点。","5-way open-source flagship comparison based on local DB RTX 4090 measurements: Qwen 3.5 27B 91 tok\u002Fs, Gemma 4 12B 89 tok\u002Fs, Qwen 3.8 Flash 180B 12 tok\u002Fs, DeepSeek V4.1 763B 3 tok\u002Fs. 27B is the sweet spot.",[306,307,309,311,313,315,317,320],{"id":4,"name":5},{"id":228,"name":308},"GLM-5.3-Flash",{"id":24,"name":310},"Qwen3.8-Flash-Next",{"id":238,"name":312},"DeepSeek-V4.1-Flash",{"id":67,"name":314},"GLM-5.3",{"id":278,"name":316},"Qwen3.5-27B",{"id":318,"name":319},63,"gemma-4-31B-it",{"id":321,"name":322},65,"gemma-4-12B-it",[324],{"id":45,"name":38},{"items":326},[]]