[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"model-phi-4-mini-38b":3},{"id":4,"slug":5,"name":6,"hf_id":7,"organization":8,"architecture":9,"architecture_config":10,"parameters":11,"parameters_label":12,"context_length":13,"license":14,"downloads":15,"likes":16,"status":17,"description":10,"huggingface_url":18,"created_at":19,"seo":20,"category":23,"family":26,"score":29,"variants":30,"quantizations":31,"requirements":58,"gpu_recommendations":138,"fitting_gpus":203,"community":239},301,"phi-4-mini-38b","Phi-4 Mini 3.8B","microsoft\u002FPhi-4-mini-instruct","microsoft","transformer",null,3840000000,"3.8B",16384,"MIT",429490,809,"published","https:\u002F\u002Fhuggingface.co\u002Fmicrosoft\u002FPhi-4-mini-instruct","2026-08-12T03:11:21+00:00",{"title":6,"description":21,"image":10,"robots":22,"canonical_url":10},"AI 模型部署决策引擎——开源 AI 的 GPU 需求、云端价格与成本分析。","index, follow",{"name":24,"slug":25},"LLM","llm",{"name":27,"slug":28},"Phi","phi",0.0096,[],[32,37,42,47,51,55],{"format":33,"bits":34,"size_factor":35,"quality_loss":36},"AWQ",4,0.263,0.03,{"format":38,"bits":39,"size_factor":40,"quality_loss":41},"FP16",16,1,0,{"format":43,"bits":44,"size_factor":45,"quality_loss":46},"FP8",8,0.5,0.01,{"format":48,"bits":34,"size_factor":49,"quality_loss":50},"GGUF",0.25,0.05,{"format":52,"bits":34,"size_factor":53,"quality_loss":54},"GPTQ",0.275,0.04,{"format":56,"bits":44,"size_factor":45,"quality_loss":57},"INT8",0.02,{"FP16":59,"FP8":82,"INT8":95,"AWQ":99,"GPTQ":112,"GGUF":125},{"minimum":60,"production":69,"recommended":75},{"scenario":61,"context_length":62,"batch_size":40,"weight_gb":63,"kv_cache_gb":45,"vram_gb":64,"ram_gb":65,"disk_gb":66,"requirement_source":67,"confidence":68},"minimum",4096,7.15,14,32,50.16,"estimated",0.95,{"scenario":70,"context_length":13,"batch_size":34,"weight_gb":63,"kv_cache_gb":71,"vram_gb":72,"ram_gb":73,"disk_gb":74,"requirement_source":67,"confidence":68},"production",64,127.47,191.21,115.16,{"scenario":76,"context_length":77,"batch_size":78,"weight_gb":63,"kv_cache_gb":44,"vram_gb":79,"ram_gb":80,"disk_gb":81,"requirement_source":67,"confidence":68},"recommended",8192,2,30.16,45.24,76.16,{"minimum":83,"production":87,"recommended":91},{"scenario":61,"context_length":62,"batch_size":40,"weight_gb":84,"kv_cache_gb":45,"vram_gb":85,"ram_gb":65,"disk_gb":86,"requirement_source":67,"confidence":68},3.58,9.47,44.58,{"scenario":70,"context_length":13,"batch_size":34,"weight_gb":84,"kv_cache_gb":71,"vram_gb":88,"ram_gb":89,"disk_gb":90,"requirement_source":67,"confidence":68},122.54,183.8,109.58,{"scenario":76,"context_length":77,"batch_size":78,"weight_gb":84,"kv_cache_gb":44,"vram_gb":92,"ram_gb":93,"disk_gb":94,"requirement_source":67,"confidence":68},25.43,38.14,70.58,{"minimum":96,"production":97,"recommended":98},{"scenario":61,"context_length":62,"batch_size":40,"weight_gb":84,"kv_cache_gb":45,"vram_gb":85,"ram_gb":65,"disk_gb":86,"requirement_source":67,"confidence":68},{"scenario":70,"context_length":13,"batch_size":34,"weight_gb":84,"kv_cache_gb":71,"vram_gb":88,"ram_gb":89,"disk_gb":90,"requirement_source":67,"confidence":68},{"scenario":76,"context_length":77,"batch_size":78,"weight_gb":84,"kv_cache_gb":44,"vram_gb":92,"ram_gb":93,"disk_gb":94,"requirement_source":67,"confidence":68},{"minimum":100,"production":104,"recommended":108},{"scenario":61,"context_length":62,"batch_size":40,"weight_gb":101,"kv_cache_gb":45,"vram_gb":102,"ram_gb":65,"disk_gb":103,"requirement_source":67,"confidence":68},1.86,7.3,41.89,{"scenario":70,"context_length":13,"batch_size":34,"weight_gb":101,"kv_cache_gb":71,"vram_gb":105,"ram_gb":106,"disk_gb":107,"requirement_source":67,"confidence":68},120.16,180.24,106.89,{"scenario":76,"context_length":77,"batch_size":78,"weight_gb":101,"kv_cache_gb":44,"vram_gb":109,"ram_gb":110,"disk_gb":111,"requirement_source":67,"confidence":68},23.15,34.73,67.89,{"minimum":113,"production":117,"recommended":121},{"scenario":61,"context_length":62,"batch_size":40,"weight_gb":114,"kv_cache_gb":45,"vram_gb":115,"ram_gb":65,"disk_gb":116,"requirement_source":67,"confidence":68},1.88,7.33,41.93,{"scenario":70,"context_length":13,"batch_size":34,"weight_gb":114,"kv_cache_gb":71,"vram_gb":118,"ram_gb":119,"disk_gb":120,"requirement_source":67,"confidence":68},120.19,180.29,106.93,{"scenario":76,"context_length":77,"batch_size":78,"weight_gb":114,"kv_cache_gb":44,"vram_gb":122,"ram_gb":123,"disk_gb":124,"requirement_source":67,"confidence":68},23.18,34.77,67.93,{"minimum":126,"production":130,"recommended":134},{"scenario":61,"context_length":62,"batch_size":40,"weight_gb":127,"kv_cache_gb":45,"vram_gb":128,"ram_gb":65,"disk_gb":129,"requirement_source":67,"confidence":68},1.92,7.38,42,{"scenario":70,"context_length":13,"batch_size":34,"weight_gb":127,"kv_cache_gb":71,"vram_gb":131,"ram_gb":132,"disk_gb":133,"requirement_source":67,"confidence":68},120.25,180.38,107,{"scenario":76,"context_length":77,"batch_size":78,"weight_gb":127,"kv_cache_gb":44,"vram_gb":135,"ram_gb":136,"disk_gb":137,"requirement_source":67,"confidence":68},23.24,34.86,68,{"recommended":139,"cheapest":157,"performance":167,"min_complexity":184,"cluster_options":199,"no_match":200,"required_vram_gb":109,"candidate_count":201,"usage":202},{"gpu_model":140,"gpu_model_id":78,"gpu_count":40,"total_vram_gb":65,"hour_price":141,"monthly_cost":142,"providers":143,"bandwidth_gbps":146,"fp16_tflops":147,"interconnect_score":148,"parallel_efficiency":40,"score":149,"tier":150,"perf_metric":151,"perf_label":152,"perf_value":153,"workload":10,"cost_metric":154},"RTX 5090",0.3222,235.21,[144,145],"vast","runpod",1792,209,40,0.6908,"cheapest","tok\u002Fs","963.4 tok\u002Fs",963.4,{"effective_value":155,"effective_unit":151,"capacity_factor":156},674.4,0.7,{"gpu_model":158,"gpu_model_id":159,"gpu_count":78,"total_vram_gb":65,"hour_price":160,"monthly_cost":161,"providers":162,"bandwidth_gbps":41,"fp16_tflops":41,"interconnect_score":148,"parallel_efficiency":163,"score":164,"tier":150,"perf_metric":151,"perf_label":165,"perf_value":41,"workload":10,"cost_metric":166},"Tesla V100",17,0.0272,39.71,[144],0.8,0.6396,"0 tok\u002Fs",{"effective_value":41,"effective_unit":151,"capacity_factor":156},{"gpu_model":168,"gpu_model_id":169,"gpu_count":44,"total_vram_gb":170,"hour_price":171,"monthly_cost":172,"providers":173,"bandwidth_gbps":175,"fp16_tflops":176,"interconnect_score":148,"parallel_efficiency":177,"score":178,"tier":179,"perf_metric":151,"perf_label":180,"perf_value":181,"workload":10,"cost_metric":182},"H200 SXM",11,1128,3.5,20440,[174],"lambda",4800,989,0.4,0.2197,"balanced","8258.1 tok\u002Fs",8258.1,{"effective_value":183,"effective_unit":151,"capacity_factor":156},5780.7,{"gpu_model":185,"gpu_model_id":186,"gpu_count":40,"total_vram_gb":187,"hour_price":188,"monthly_cost":189,"providers":190,"bandwidth_gbps":192,"fp16_tflops":193,"interconnect_score":148,"parallel_efficiency":40,"score":194,"tier":150,"perf_metric":151,"perf_label":195,"perf_value":196,"workload":10,"cost_metric":197},"RTX 3090",3,24,0.0678,49.49,[144,191],"tensordock",936,71,0.6423,"503.2 tok\u002Fs",503.2,{"effective_value":198,"effective_unit":151,"capacity_factor":156},352.2,[],false,25,"value",[204,208,216,223,231],{"name":185,"vram_gb":187,"cheapest_hour":188,"cost":205,"provider":144,"estimated_tps":196},{"hourly":188,"daily":206,"monthly":189,"yearly":207},1.63,593.93,{"name":209,"vram_gb":187,"cheapest_hour":210,"cost":211,"provider":144,"estimated_tps":215},"RTX 4090",0.1344,{"hourly":210,"daily":212,"monthly":213,"yearly":214},3.23,98.11,1177.34,541.9,{"name":217,"vram_gb":187,"cheapest_hour":218,"cost":219,"provider":144,"estimated_tps":215},"RTX 4090D",0.1602,{"hourly":218,"daily":220,"monthly":221,"yearly":222},3.84,116.95,1403.35,{"name":224,"vram_gb":187,"cheapest_hour":225,"cost":226,"provider":144,"estimated_tps":230},"L4",0.2015,{"hourly":225,"daily":227,"monthly":228,"yearly":229},4.84,147.1,1765.14,161.3,{"name":232,"vram_gb":187,"cheapest_hour":233,"cost":234,"provider":144,"estimated_tps":238},"A10",0.2409,{"hourly":233,"daily":235,"monthly":236,"yearly":237},5.78,175.86,2110.28,322.6,{"posts_count":41,"benchmarks_count":41}]