[{"data":1,"prerenderedAt":1838},["ShallowReactive",2],{"blog-post-nl-\u002Fblog\u002Fpaiton-qwen38-3bit-accuracy-prefix-caching-mtp":3,"blog-posts-sidebar-nl":1300},{"id":4,"title":5,"body":6,"categories":1281,"date":1286,"description":1287,"extension":1288,"heading":1289,"image":1290,"meta":1291,"navigation":776,"originalUrl":1292,"path":1293,"seo":1294,"slug":1295,"socialImage":1296,"stem":1297,"updated":1298,"video":1298,"__hash__":1299},"blogNl\u002Fblog\u002Fpaiton-qwen38-3bit-accuracy-prefix-caching-mtp.md","Qwen3.8 Flash Next: 3-bit dicht bij 16-bit nauwkeurigheid, met MTP-prefixcaching",{"type":7,"value":8,"toc":1268},"minimark",[9,16,48,63,68,87,91,96,193,214,231,236,283,302,306,323,327,336,339,344,427,440,460,483,498,502,511,514,519,590,616,620,638,654,658,673,690,694,715,726,872,900,910,929,944,948,971,984,1264],[10,11,12],"p",{},[13,14,15],"em",{},"Illustratie: met AI gemaakt redactioneel beeld, geen hardwarefoto of benchmarkopname.",[10,17,18,19,23,24,27,28,40],{},"Betere antwoorden, minder wachten tussen beurten. Onze Paiton-release van 24 oktober brengt ",[20,21,22],"strong",{},"nieuwe 3-bit gewichten voor Qwen3.8 Flash Next die op drie taakbenchmarks dicht bij de 16-bit referentie liggen",", plus native prefixcaching met multi-token prediction (MTP) voor lange gesprekken op ",[20,25,26],{},"twee Radeon AI PRO R9700-kaarten",". Een afzonderlijke nauwkeurigheidsupdate verbetert ook ons 27B-model op één kaart.",[29,30,31],"sup",{},[32,33,39],"a",{"href":34,"ariaDescribedBy":35,"dataFootnoteRef":37,"id":38},"#user-content-fn-threebit",[36],"footnote-label","","user-content-fnref-threebit","1",[29,41,42],{},[32,43,47],{"href":44,"ariaDescribedBy":45,"dataFootnoteRef":37,"id":46},"#user-content-fn-twentyseven",[36],"user-content-fnref-twentyseven","2",[10,49,50,51,54,55],{},"De nieuwste 3-bit cachemodus hervat een herhaalde ",[20,52,53],{},"prompt van 200K tokens in 0,68 seconde",", tegenover 25,81 seconden zonder cachehit, met speculatieve generatie aan. Dit zijn cachemetingen, niet de wachttijd voor een nieuw document.",[29,56,57],{},[32,58,62],{"href":59,"ariaDescribedBy":60,"dataFootnoteRef":37,"id":61},"#user-content-fn-bench",[36],"user-content-fnref-bench","3",[64,65,67],"h2",{"id":66},"nauwkeurigheid-vergelijkbare-taakscores-in-3-bit","Nauwkeurigheid: vergelijkbare taakscores in 3-bit",[10,69,70,71,75,76,80,81],{},"De nieuwe gewichten verbeteren de kalibratie achter onze ",[32,72,74],{"href":73},"\u002Fnl\u002Fblog\u002Fpaiton-qwen38-flash-next-radeon-ai-pro-r9700","oorspronkelijke 3-bit Flash Next-release",", zonder het formaat, de grootte of de kernels te veranderen. Hieronder staan ze naast de ",[32,77,79],{"href":78},"\u002Fnl\u002Fblog\u002Fpaiton-qwen38-flash-next-4bit-radeon-ai-pro-r9700","4-bit release"," en het oorspronkelijke 16-bit model.",[29,82,83],{},[32,84,39],{"href":34,"ariaDescribedBy":85,"dataFootnoteRef":37,"id":86},[36],"user-content-fnref-threebit-2",[88,89],"qwen-accuracy-mtp-figure",{"kind":90},"accuracy",[10,92,93],{},[13,94,95],{},"Nauwkeurigheid; hoger is beter. Teksttaken draaiden via de uitgebrachte serverconfiguraties, niet de native doorvoeropstelling hieronder. AI2D emuleert de berekeningen tijdens serving en staat apart. De 3-bit resultaten zijn van 24 oktober; de 4-bit en 16-bit referenties zijn van 12 oktober.",[97,98,99,122],"table",{},[100,101,102],"thead",{},[103,104,105,109,113,116,119],"tr",{},[106,107,108],"th",{},"Taak",[106,110,112],{"align":111},"right","Vragen",[106,114,115],{"align":111},"16-bit",[106,117,118],{"align":111},"Nieuwe 3-bit",[106,120,121],{"align":111},"4-bit",[123,124,125,143,159,176],"tbody",{},[103,126,127,131,134,137,140],{},[128,129,130],"td",{},"GSM8K, rekenen",[128,132,133],{"align":111},"1.319",[128,135,136],{"align":111},"96,9%",[128,138,139],{"align":111},"97,4%",[128,141,142],{"align":111},"96,8%",[103,144,145,148,151,154,157],{},[128,146,147],{},"HumanEval, Python pass@1",[128,149,150],{"align":111},"164",[128,152,153],{"align":111},"96,3%",[128,155,156],{"align":111},"94,5%",[128,158,153],{"align":111},[103,160,161,164,167,170,173],{},[128,162,163],{},"MMLU-Pro, kennis",[128,165,166],{"align":111},"1.400",[128,168,169],{"align":111},"84,7%",[128,171,172],{"align":111},"83,6%",[128,174,175],{"align":111},"83,7%",[103,177,178,181,184,187,190],{},[128,179,180],{},"AI2D, wetenschappelijke diagrammen",[128,182,183],{"align":111},"600",[128,185,186],{"align":111},"85,3%",[128,188,189],{"align":111},"81,8%",[128,191,192],{"align":111},"Niet gerapporteerd",[10,194,195,196,199,200,206],{},"Geen van de drie teksttaken verschilt statistisch significant van 16-bit in de gepaarde toetsen. Dat ",[20,197,198],{},"bewijst geen gelijkwaardigheid",": HumanEval ligt 1,8 procentpunt lager, met een gepaard 95%-interval van −5,0 tot +1,3 punten. De 16-bit referentie zelf veranderde met maximaal 1,2 punten tussen twee reeksen seeds.",[29,201,202],{},[32,203,39],{"href":34,"ariaDescribedBy":204,"dataFootnoteRef":37,"id":205},[36],"user-content-fnref-threebit-3",[29,207,208],{},[32,209,213],{"href":210,"ariaDescribedBy":211,"dataFootnoteRef":37,"id":212},"#user-content-fn-fourbit",[36],"user-content-fnref-fourbit","4",[10,215,216,217,223],{},"Beeldbegrip blijft achter op de referentie. De AI2D-evaluatie toont niet aan dat de native worker beeldinvoer ondersteunt: de uitgebrachte 3-bit modi verwerken alleen tekst.",[29,218,219],{},[32,220,39],{"href":34,"ariaDescribedBy":221,"dataFootnoteRef":37,"id":222},[36],"user-content-fnref-threebit-4",[29,224,225],{},[32,226,230],{"href":227,"ariaDescribedBy":228,"dataFootnoteRef":37,"id":229},"#user-content-fn-release",[36],"user-content-fnref-release","5",[232,233,235],"h3",{"id":234},"hoe-dicht-liggen-de-tokenvoorspellingen-bij-elkaar","Hoe dicht liggen de tokenvoorspellingen bij elkaar?",[97,237,238,249],{},[100,239,240],{},[103,241,242,245,247],{},[106,243,244],{},"Afzonderlijke evaluatie tegenover 16-bit",[106,246,118],{"align":111},[106,248,121],{"align":111},[123,250,251,262,273],{},[103,252,253,256,259],{},[128,254,255],{},"KL-divergentie over de volledige woordenschat, lager is beter",[128,257,258],{"align":111},"0,055",[128,260,261],{"align":111},"0,031",[103,263,264,267,270],{},[128,265,266],{},"Relatieve toename van negatieve log-likelihood bij lange documenten",[128,268,269],{"align":111},"+0,35%",[128,271,272],{"align":111},"+0,16%",[103,274,275,278,281],{},[128,276,277],{},"Naald-in-een-hooibergtest bij 131.072 tokens",[128,279,280],{"align":111},"40 \u002F 40",[128,282,280],{"align":111},[10,284,285,286,289,290,296],{},"KL meet verschillen in de kansverdeling voor het volgende token; nul betekent identieke verdelingen. Negatieve log-likelihood meet hoe goed voorspellingen bij waargenomen tekst passen. ",[20,287,288],{},"Dit zijn relatieve NLL-veranderingen, geen veranderingen in perplexiteit."," De verdelingscontroles emuleren de berekeningen tijdens serving; de afzonderlijke terugvindtests gebruiken het geserveerde model. Geen van beide vervangt een brede taakevaluatie.",[29,291,292],{},[32,293,39],{"href":34,"ariaDescribedBy":294,"dataFootnoteRef":37,"id":295},[36],"user-content-fnref-threebit-5",[29,297,298],{},[32,299,213],{"href":210,"ariaDescribedBy":300,"dataFootnoteRef":37,"id":301},[36],"user-content-fnref-fourbit-2",[232,303,305],{"id":304},"wat-veranderde-aan-de-gewichten","Wat veranderde aan de gewichten",[10,307,308,309,312,313,316,317],{},"Het recept combineert afzonderlijke kalibratiestatistieken voor de uitvoerprojectie van elke expert, een gevarieerdere kalibratieset van ongeveer ",[20,310,311],{},"vier miljoen tokens",", en behouden ",[20,314,315],{},"8-bit gedeelde lagen"," rond de 3-bit experts. Alleen de afronding verandert; het gepakte formaat en de runtimekernels blijven gelijk.",[29,318,319],{},[32,320,39],{"href":34,"ariaDescribedBy":321,"dataFootnoteRef":37,"id":322},[36],"user-content-fnref-threebit-6",[64,324,326],{"id":325},"prefixcaching-terwijl-mtp-aan-blijft","Prefixcaching terwijl MTP aan blijft",[10,328,329,330],{},"Codeeragents sturen bij elke beurt het hele gesprek opnieuw. Prefixcaching hergebruikt het onveranderde deel; MTP stelt meerdere volgende tokens voor die het doelmodel controleert. De native worker combineert nu beide. Hij voert de modelstap uit zonder PyTorch op de GPU, terwijl vLLM de API en planning blijft verzorgen.",[29,331,332],{},[32,333,230],{"href":227,"ariaDescribedBy":334,"dataFootnoteRef":37,"id":335},[36],"user-content-fnref-release-2",[88,337],{"kind":338},"prefix-cache",[10,340,341],{},[13,342,343],{},"Tijd tot het eerste token in afzonderlijke tests met herhaalde prompts; lager is beter. Een hit vereist een onveranderde prefix die nog in de GPU- of RAM-cache staat. Dit zijn geen prefilldoorvoermetingen met een lege cache.",[97,345,346,359],{},[100,347,348],{},[103,349,350,353,356],{},[106,351,352],{},"Configuratie en herhaalde prompt",[106,354,355],{"align":111},"Zonder cachehit",[106,357,358],{"align":111},"Met cachehit",[123,360,361,372,383,394,405,416],{},[103,362,363,366,369],{},[128,364,365],{},"4-bit native MTP-cache, 32K",[128,367,368],{"align":111},"5,25 s",[128,370,371],{"align":111},"0,69 s",[103,373,374,377,380],{},[128,375,376],{},"4-bit native MTP-cache, 90K",[128,378,379],{"align":111},"14,68 s",[128,381,382],{"align":111},"0,78 s",[103,384,385,388,391],{},[128,386,387],{},"4-bit native MTP-cache, 200K",[128,389,390],{"align":111},"33,62 s",[128,392,393],{"align":111},"0,85 s",[103,395,396,399,402],{},[128,397,398],{},"4-bit inactieve sessie hersteld uit RAM, 145K–200K",[128,400,401],{"align":111},"21–30 s",[128,403,404],{"align":111},"0,98–1,17 s",[103,406,407,410,413],{},[128,408,409],{},"Nieuwe 3-bit native MTP-cache, 90K",[128,411,412],{"align":111},"11,09 s",[128,414,415],{"align":111},"0,60 s",[103,417,418,421,424],{},[128,419,420],{},"Nieuwe 3-bit native MTP-cache, 200K",[128,422,423],{"align":111},"25,81 s",[128,425,426],{"align":111},"0,68 s",[10,428,429,430,433,434],{},"Afzonderlijk verwerkte testverzoeken leveren identieke uitvoer met en zonder cachehit; gelijktijdige verwerking kan op beide paden tokens veranderen. Een optionele RAM-laag bewaart verwijderde sessies en past zich aan het beschikbare geheugen aan. De 4-bit cacheconfiguratie genereert ",[20,431,432],{},"206,9 tokens\u002Fs met die laag aan",".",[29,435,436],{},[32,437,62],{"href":59,"ariaDescribedBy":438,"dataFootnoteRef":37,"id":439},[36],"user-content-fnref-bench-2",[10,441,442,443,449,450,453,454],{},"De nieuwere modus ",[20,444,445],{},[446,447,448],"code",{},"3bit-long-cache-mtp-native"," brengt dit naar een ",[20,451,452],{},"contextvenster van 262.144 tokens",": 213,5 tokens\u002Fs met de RAM-laag aan, 250,0 zonder. De cache- en generatiecijfers komen uit afzonderlijke validatiemetingen, niet uit herhalingen van de releasevergelijking hieronder. Prompt, opmaak en antwoord moeten samen in het venster passen.",[29,455,456],{},[32,457,62],{"href":59,"ariaDescribedBy":458,"dataFootnoteRef":37,"id":459},[36],"user-content-fnref-bench-3",[10,461,462,463,468,469,472,473,476,477],{},"Voor kleinere hosts ondersteunt ",[20,464,465],{},[446,466,467],{},"--mode decode --fn-worker native --prefix-caching"," ",[20,470,471],{},"64 GB RAM",", een venster van 98.304 tokens en ongeveer 101 tokens\u002Fs, ",[20,474,475],{},"zonder MTP",". Dit is een andere modus, niet de configuratie achter het resultaat van 213,5 tokens\u002Fs.",[29,478,479],{},[32,480,230],{"href":227,"ariaDescribedBy":481,"dataFootnoteRef":37,"id":482},[36],"user-content-fnref-release-3",[10,484,485,486,492],{},"In één test dicht bij de capaciteit, met 2,05 miljoen tokens aan inactieve sessies, werden verwachte RAM-hits toch opnieuw ingelezen. De uitvoer bleef correct; het probleem staat nog open. Ook eindigt het 3-bit model in geteste sessies vaker te vroeg voorbij ongeveer 28K context. De openbare handleiding geeft voor lange codegesprekken de voorkeur aan 4-bit.",[29,487,488],{},[32,489,62],{"href":59,"ariaDescribedBy":490,"dataFootnoteRef":37,"id":491},[36],"user-content-fnref-bench-4",[29,493,494],{},[32,495,39],{"href":34,"ariaDescribedBy":496,"dataFootnoteRef":37,"id":497},[36],"user-content-fnref-threebit-7",[64,499,501],{"id":500},"generatiesnelheid-op-twee-workstationkaarten","Generatiesnelheid op twee workstationkaarten",[10,503,504,505],{},"De nauwkeurigheidsupdate verandert de gewichtswaarden, niet de hoeveelheid gelezen data of de kernels. De gepubliceerde native-workerdoorvoer blijft in hetzelfde prestatiebereik.",[29,506,507],{},[32,508,62],{"href":59,"ariaDescribedBy":509,"dataFootnoteRef":37,"id":510},[36],"user-content-fnref-bench-5",[88,512],{"kind":513},"throughput",[10,515,516],{},[13,517,518],{},"Twee R9700-kaarten, native worker, prefixcaching uit. Hoger is beter. Gewogen generatie met één verzoek en gezamenlijke doorvoer bij gelijktijdige verzoeken zijn verschillende workloads, geen schaalvergelijking per gebruiker.",[97,520,521,533],{},[100,522,523],{},[103,524,525,528,531],{},[106,526,527],{},"Meting",[106,529,530],{"align":111},"3-bit",[106,532,121],{"align":111},[123,534,535,546,557,568,579],{},[103,536,537,540,543],{},[128,538,539],{},"BetterBench gewogen generatie, één verzoek",[128,541,542],{"align":111},"258,7 tokens\u002Fs",[128,544,545],{"align":111},"245,3 tokens\u002Fs",[103,547,548,551,554],{},[128,549,550],{},"BetterBench, acht verzoeken, gezamenlijk",[128,552,553],{"align":111},"687,5 tokens\u002Fs",[128,555,556],{"align":111},"566,7 tokens\u002Fs",[103,558,559,562,565],{},[128,560,561],{},"Lezen van een nominale prompt van 64K",[128,563,564],{"align":111},"8.433 invoertokens\u002Fs",[128,566,567],{"align":111},"7.788 invoertokens\u002Fs",[103,569,570,573,576],{},[128,571,572],{},"LocalMaxxing, één gebruiker",[128,574,575],{"align":111},"220,3 tokens\u002Fs",[128,577,578],{"align":111},"200,5 tokens\u002Fs",[103,580,581,584,587],{},[128,582,583],{},"LocalMaxxing, acht gebruikers, gezamenlijk",[128,585,586],{"align":111},"781,3 tokens\u002Fs",[128,588,589],{"align":111},"680,2 tokens\u002Fs",[10,591,592,595,596,599,600,603,604,610],{},[20,593,594],{},"Dit zijn resultaten met de expliciet gekozen native worker."," Gewone ",[446,597,598],{},"decode","- en ",[446,601,602],{},"4bit","-commando's kiezen nog steeds de torch worker, tenzij anders gevraagd. De cachemodi en hun RAM-lagen zijn niet opgenomen in deze tabel.",[29,605,606],{},[32,607,230],{"href":227,"ariaDescribedBy":608,"dataFootnoteRef":37,"id":609},[36],"user-content-fnref-release-4",[29,611,612],{},[32,613,62],{"href":59,"ariaDescribedBy":614,"dataFootnoteRef":37,"id":615},[36],"user-content-fnref-bench-6",[64,617,619],{"id":618},"een-tweede-nauwkeurigheidsupdate-voor-qwen38-27b","Een tweede nauwkeurigheidsupdate voor Qwen3.8 27B",[10,621,622,623,627,628,631,632],{},"Onze ",[32,624,626],{"href":625},"\u002Fnl\u002Fblog\u002Fpaiton-qwen38-w3a4-radeon-ai-pro-r9700","3-bit 27B-release"," draait op één R9700. Deze update houdt de 3-bit codes vast en traint hun schaalfactoren verder richting het 16-bit model, met ",[20,629,630],{},"4,2 miljoen nieuwe tokens"," die losstaan van eerdere kalibratie en gerapporteerde benchmarks.",[29,633,634],{},[32,635,47],{"href":44,"ariaDescribedBy":636,"dataFootnoteRef":37,"id":637},[36],"user-content-fnref-twentyseven-2",[10,639,640,641,644,645,433,648],{},"De geserveerde KL daalt met ",[20,642,643],{},"3,0%",", met een 95%-interval van 0,4% tot 6,4%. Voorspellingen verbeteren in de lengtecategorieën van 8K–32K; de 131K-controle vindt geen verslechtering, met winst vanaf 64K. Formaat, grootte en kernelkosten blijven gelijk; gemeten doorvoer kan variëren met de acceptatie van speculatieve tokens. Nieuwe taakcontroles tonen geen significante verandering, maar zijn ",[20,646,647],{},"evaluaties van alleen de gewichten, geen nieuwe taakmetingen van het geserveerde model",[29,649,650],{},[32,651,47],{"href":44,"ariaDescribedBy":652,"dataFootnoteRef":37,"id":653},[36],"user-content-fnref-twentyseven-3",[64,655,657],{"id":656},"zo-lees-je-de-metingen","Zo lees je de metingen",[10,659,660,661,667],{},"De teksttaken voor Flash Next gebruiken thinking aan, temperature 1,0, top-p 0,95, top-k 20 en een antwoordlimiet van 16.384 tokens. Prompts en seeds per vraag zijn gelijk; antwoorden op de limiet tellen als fout. GSM8K en HumanEval gebruiken hun volledige testsets; MMLU-Pro gebruikt 100 vragen uit elk van 14 vakgebieden. De oorspronkelijke 16-bit gewichten draaiden tweemaal op een grotere AMD-server. Vergelijkingen gebruiken gepaarde betrouwbaarheidsintervallen en exacte McNemar-toetsen.",[29,662,663],{},[32,664,39],{"href":34,"ariaDescribedBy":665,"dataFootnoteRef":37,"id":666},[36],"user-content-fnref-threebit-8",[29,668,669],{},[32,670,213],{"href":210,"ariaDescribedBy":671,"dataFootnoteRef":37,"id":672},[36],"user-content-fnref-fourbit-3",[10,674,675,676,679,680,683,684],{},"De snelheidsmetingen gebruiken ",[20,677,678],{},"BetterBench 0.6.0, standaardprofiel",", één run per configuratie, en ",[20,681,682],{},"LocalMaxxing v0.1.54",", met na opwarming de run met de mediane eengebruikersdoorvoer uit drie runs. Dit zijn onze metingen, geen onafhankelijke reproducties.",[29,685,686],{},[32,687,62],{"href":59,"ariaDescribedBy":688,"dataFootnoteRef":37,"id":689},[36],"user-content-fnref-bench-7",[64,691,693],{"id":692},"probeer-de-release","Probeer de release",[10,695,696,697,700,701,704,705,708,709],{},"Gebruik ",[20,698,699],{},"Linux x86-64, twee R9700-kaarten, een ROCm 10-hostdriver, Docker met GPU-toegang en Python 3.10+",". Snelle native 3-bit en 3-bit MTP-caching vragen ongeveer ",[20,702,703],{},"128 GB+ RAM en 60 GiB vrije GTT","; native 4-bit vraagt ongeveer een ",[20,706,707],{},"GTT-limiet van 98 GiB",". De voorafgaande controle toetst de echte host. Reken op 116 GB voor 3-bit gewichten of 133 GB voor 4-bit, plus de native image van ongeveer 56 GB.",[29,710,711],{},[32,712,230],{"href":227,"ariaDescribedBy":713,"dataFootnoteRef":37,"id":714},[36],"user-content-fnref-release-5",[10,716,717,718],{},"Gebruik een nieuwe checkout als je lokale pluginwijzigingen hebt. De vastgezette launcher kiest zijn runtime- en modelrevisies, downloadt de bijgewerkte 3-bit gewichten in een nieuwe map en laat de oude download intact.",[29,719,720],{},[32,721,725],{"href":722,"ariaDescribedBy":723,"dataFootnoteRef":37,"id":724},"#user-content-fn-launcher",[36],"user-content-fnref-launcher","6",[727,728,732],"pre",{"className":729,"code":730,"language":731,"meta":37,"style":37},"language-bash shiki shiki-themes github-light github-dark","git clone https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin.git\ncd paiton-vllm-plugin\ngit checkout 4882e36e455c1e657175db910367c992837c72a5\n\npython3 -m venv .venv\n. .venv\u002Fbin\u002Factivate\npython3 -m pip install huggingface_hub\n\npython3 models\u002FQwen3.8-Flash-Next\u002Flaunch-flashnext.py \\\n  --mode 3bit-long-cache-mtp-native --host 127.0.0.1 --dry-run\npython3 models\u002FQwen3.8-Flash-Next\u002Flaunch-flashnext.py \\\n  --mode 3bit-long-cache-mtp-native --host 127.0.0.1\n","bash",[446,733,734,750,760,771,778,793,801,817,822,833,851,860],{"__ignoreMap":37},[735,736,739,743,747],"span",{"class":737,"line":738},"line",1,[735,740,742],{"class":741},"sScJk","git",[735,744,746],{"class":745},"sZZnC"," clone",[735,748,749],{"class":745}," https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin.git\n",[735,751,753,757],{"class":737,"line":752},2,[735,754,756],{"class":755},"sj4cs","cd",[735,758,759],{"class":745}," paiton-vllm-plugin\n",[735,761,763,765,768],{"class":737,"line":762},3,[735,764,742],{"class":741},[735,766,767],{"class":745}," checkout",[735,769,770],{"class":745}," 4882e36e455c1e657175db910367c992837c72a5\n",[735,772,774],{"class":737,"line":773},4,[735,775,777],{"emptyLinePlaceholder":776},true,"\n",[735,779,781,784,787,790],{"class":737,"line":780},5,[735,782,783],{"class":741},"python3",[735,785,786],{"class":755}," -m",[735,788,789],{"class":745}," venv",[735,791,792],{"class":745}," .venv\n",[735,794,796,798],{"class":737,"line":795},6,[735,797,433],{"class":755},[735,799,800],{"class":745}," .venv\u002Fbin\u002Factivate\n",[735,802,804,806,808,811,814],{"class":737,"line":803},7,[735,805,783],{"class":741},[735,807,786],{"class":755},[735,809,810],{"class":745}," pip",[735,812,813],{"class":745}," install",[735,815,816],{"class":745}," huggingface_hub\n",[735,818,820],{"class":737,"line":819},8,[735,821,777],{"emptyLinePlaceholder":776},[735,823,825,827,830],{"class":737,"line":824},9,[735,826,783],{"class":741},[735,828,829],{"class":745}," models\u002FQwen3.8-Flash-Next\u002Flaunch-flashnext.py",[735,831,832],{"class":755}," \\\n",[735,834,836,839,842,845,848],{"class":737,"line":835},10,[735,837,838],{"class":755},"  --mode",[735,840,841],{"class":745}," 3bit-long-cache-mtp-native",[735,843,844],{"class":755}," --host",[735,846,847],{"class":755}," 127.0.0.1",[735,849,850],{"class":755}," --dry-run\n",[735,852,854,856,858],{"class":737,"line":853},11,[735,855,783],{"class":741},[735,857,829],{"class":745},[735,859,832],{"class":755},[735,861,863,865,867,869],{"class":737,"line":862},12,[735,864,838],{"class":755},[735,866,841],{"class":745},[735,868,844],{"class":755},[735,870,871],{"class":755}," 127.0.0.1\n",[10,873,874,875,878,879,882,883,886,887,890,891,433,894],{},"Gebruik voor 4-bit MTP-caching ",[446,876,877],{},"bash models\u002FQwen3.8-Flash-Next\u002Frun-flashnext-4bit.sh"," met ",[446,880,881],{},"--fn-worker native --prefix-caching --host 127.0.0.1",". Wacht op ",[446,884,885],{},"launcher: READY","; de API staat op ",[446,888,889],{},"http:\u002F\u002F127.0.0.1:18982\u002Fv1",", met modelnaam ",[446,892,893],{},"Qwen3.8-Flash-Next",[29,895,896],{},[32,897,230],{"href":227,"ariaDescribedBy":898,"dataFootnoteRef":37,"id":899},[36],"user-content-fnref-release-6",[10,901,902,903,909],{},"Volg voor 27B de ",[32,904,908],{"href":905,"rel":906},"https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin\u002Fblob\u002F4882e36e455c1e657175db910367c992837c72a5\u002Fmodels\u002FQwen3.8-MXFP4-DFlash2\u002FREADME.md",[907],"nofollow","vastgezette installatie- en upgradehandleiding",". Download de nieuwe gewichtsrevisie expliciet; alleen de launcher bijwerken vernieuwt die gewichten niet.",[10,911,912,913,918,919,922,923],{},"De server luistert standaard op alle netwerkinterfaces zonder API-sleutel. Gebruik ",[20,914,915],{},[446,916,917],{},"--host 127.0.0.1"," voor lokaal gebruik. Bescherm toegang voor externe clients met een firewall en ",[446,920,921],{},"PAITON_API_KEY","; zet een sleutel niet rechtstreeks in commandoargumenten. Stop de draaiende server voordat je een andere modus kiest.",[29,924,925],{},[32,926,230],{"href":227,"ariaDescribedBy":927,"dataFootnoteRef":37,"id":928},[36],"user-content-fnref-release-7",[10,930,931,932,937,938],{},"Native modi weigeren gestructureerde uitvoer, strikte\u002Fverplichte toolaanroepen en andere API-opties; gewone automatische toolaanroepen werken. Lees de ",[32,933,936],{"href":934,"rel":935},"https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin\u002Fblob\u002F4882e36e455c1e657175db910367c992837c72a5\u002Fmodels\u002FQwen3.8-Flash-Next\u002FREADME.md#native-worker-opt-in",[907],"compatibiliteitsinformatie"," voordat je een agent aansluit.",[29,939,940],{},[32,941,230],{"href":227,"ariaDescribedBy":942,"dataFootnoteRef":37,"id":943},[36],"user-content-fnref-release-8",[64,945,947],{"id":946},"credits-en-licenties","Credits en licenties",[10,949,950,951,954,955,958,959,965],{},"Qwen levert de modellen, vLLM het servingframework, en BetterBench en LocalMaxxing de benchmarktools. Flash Next-gewichten gebruiken ",[20,952,953],{},"Qwen Community License 1.0",", met voorwaarden voor commercieel gebruik; de 27B-gewichten gebruiken ",[20,956,957],{},"Apache 2.0",". Lees vóór inzet de modellicenties en naamsvermeldingen voor kalibratiedata.",[29,960,961],{},[32,962,39],{"href":34,"ariaDescribedBy":963,"dataFootnoteRef":37,"id":964},[36],"user-content-fnref-threebit-9",[29,966,967],{},[32,968,47],{"href":44,"ariaDescribedBy":969,"dataFootnoteRef":37,"id":970},[36],"user-content-fnref-twentyseven-4",[10,972,973,974,978,979,983],{},"Bekijk ",[32,975,977],{"href":976},"\u002Fnl\u002Fproducts\u002Fpaiton","Paiton"," voor de runtime, of ",[32,980,982],{"href":981},"\u002Fnl\u002Fproducts\u002Fsovereign-ai","soevereine AI"," voor een lokaal beheerde implementatie. De vastgezette bronnen hieronder beschrijven kalibratie, modi en reproduceerbaarheid.",[985,986,989,994],"section",{"className":987,"dataFootnotes":37},[988],"footnotes",[64,990,993],{"className":991,"id":36},[992],"sr-only","Footnotes",[995,996,997,1073,1107,1162,1189,1251],"ol",{},[998,999,1001,1006,1007,468,1014,468,1021,468,1028,468,1035,468,1042,468,1049,468,1057,468,1065],"li",{"id":1000},"user-content-fn-threebit",[32,1002,1005],{"href":1003,"rel":1004},"https:\u002F\u002Fhuggingface.co\u002FEliovpAI\u002FQwen3.8-Flash-Next-W3A8-Paiton-RDNA4\u002Fblob\u002Fa100b3aee53b2cf2c9636650387936aab22dd5d7\u002FREADME.md",[907],"Flash Next W3A8-modelkaart, taaknauwkeurigheid en kalibratie, vastgezette revisie",". ",[32,1008,1013],{"href":1009,"ariaLabel":1010,"className":1011,"dataFootnoteBackref":37},"#user-content-fnref-threebit","Back to reference 1",[1012],"data-footnote-backref","↩",[32,1015,1013,1019],{"href":1016,"ariaLabel":1017,"className":1018,"dataFootnoteBackref":37},"#user-content-fnref-threebit-2","Back to reference 1-2",[1012],[29,1020,47],{},[32,1022,1013,1026],{"href":1023,"ariaLabel":1024,"className":1025,"dataFootnoteBackref":37},"#user-content-fnref-threebit-3","Back to reference 1-3",[1012],[29,1027,62],{},[32,1029,1013,1033],{"href":1030,"ariaLabel":1031,"className":1032,"dataFootnoteBackref":37},"#user-content-fnref-threebit-4","Back to reference 1-4",[1012],[29,1034,213],{},[32,1036,1013,1040],{"href":1037,"ariaLabel":1038,"className":1039,"dataFootnoteBackref":37},"#user-content-fnref-threebit-5","Back to reference 1-5",[1012],[29,1041,230],{},[32,1043,1013,1047],{"href":1044,"ariaLabel":1045,"className":1046,"dataFootnoteBackref":37},"#user-content-fnref-threebit-6","Back to reference 1-6",[1012],[29,1048,725],{},[32,1050,1013,1054],{"href":1051,"ariaLabel":1052,"className":1053,"dataFootnoteBackref":37},"#user-content-fnref-threebit-7","Back to reference 1-7",[1012],[29,1055,1056],{},"7",[32,1058,1013,1062],{"href":1059,"ariaLabel":1060,"className":1061,"dataFootnoteBackref":37},"#user-content-fnref-threebit-8","Back to reference 1-8",[1012],[29,1063,1064],{},"8",[32,1066,1013,1070],{"href":1067,"ariaLabel":1068,"className":1069,"dataFootnoteBackref":37},"#user-content-fnref-threebit-9","Back to reference 1-9",[1012],[29,1071,1072],{},"9",[998,1074,1076,1006,1081,468,1086,468,1093,468,1100],{"id":1075},"user-content-fn-twentyseven",[32,1077,1080],{"href":1078,"rel":1079},"https:\u002F\u002Fhuggingface.co\u002FEliovpAI\u002FQwen3.8-27B-W3Rot-INT3-Paiton-RDNA4\u002Fblob\u002F383fa4b221a992fc2437500328056ef05355c8eb\u002FREADME.md",[907],"27B 3-bit modelkaart en tweede nauwkeurigheidsupdate, vastgezette revisie",[32,1082,1013],{"href":1083,"ariaLabel":1084,"className":1085,"dataFootnoteBackref":37},"#user-content-fnref-twentyseven","Back to reference 2",[1012],[32,1087,1013,1091],{"href":1088,"ariaLabel":1089,"className":1090,"dataFootnoteBackref":37},"#user-content-fnref-twentyseven-2","Back to reference 2-2",[1012],[29,1092,47],{},[32,1094,1013,1098],{"href":1095,"ariaLabel":1096,"className":1097,"dataFootnoteBackref":37},"#user-content-fnref-twentyseven-3","Back to reference 2-3",[1012],[29,1099,62],{},[32,1101,1013,1105],{"href":1102,"ariaLabel":1103,"className":1104,"dataFootnoteBackref":37},"#user-content-fnref-twentyseven-4","Back to reference 2-4",[1012],[29,1106,213],{},[998,1108,1110,1006,1115,468,1120,468,1127,468,1134,468,1141,468,1148,468,1155],{"id":1109},"user-content-fn-bench",[32,1111,1114],{"href":1112,"rel":1113},"https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin\u002Fblob\u002F4882e36e455c1e657175db910367c992837c72a5\u002Fmodels\u002FQwen3.8-Flash-Next\u002FBENCHMARKS.md",[907],"Flash Next-snelheidsmetingen, cachevalidatie en methodologie, vastgezette release",[32,1116,1013],{"href":1117,"ariaLabel":1118,"className":1119,"dataFootnoteBackref":37},"#user-content-fnref-bench","Back to reference 3",[1012],[32,1121,1013,1125],{"href":1122,"ariaLabel":1123,"className":1124,"dataFootnoteBackref":37},"#user-content-fnref-bench-2","Back to reference 3-2",[1012],[29,1126,47],{},[32,1128,1013,1132],{"href":1129,"ariaLabel":1130,"className":1131,"dataFootnoteBackref":37},"#user-content-fnref-bench-3","Back to reference 3-3",[1012],[29,1133,62],{},[32,1135,1013,1139],{"href":1136,"ariaLabel":1137,"className":1138,"dataFootnoteBackref":37},"#user-content-fnref-bench-4","Back to reference 3-4",[1012],[29,1140,213],{},[32,1142,1013,1146],{"href":1143,"ariaLabel":1144,"className":1145,"dataFootnoteBackref":37},"#user-content-fnref-bench-5","Back to reference 3-5",[1012],[29,1147,230],{},[32,1149,1013,1153],{"href":1150,"ariaLabel":1151,"className":1152,"dataFootnoteBackref":37},"#user-content-fnref-bench-6","Back to reference 3-6",[1012],[29,1154,725],{},[32,1156,1013,1160],{"href":1157,"ariaLabel":1158,"className":1159,"dataFootnoteBackref":37},"#user-content-fnref-bench-7","Back to reference 3-7",[1012],[29,1161,1056],{},[998,1163,1165,1006,1170,468,1175,468,1182],{"id":1164},"user-content-fn-fourbit",[32,1166,1169],{"href":1167,"rel":1168},"https:\u002F\u002Fhuggingface.co\u002FEliovpAI\u002FQwen3.8-Flash-Next-W4A8-Paiton-RDNA4\u002Fblob\u002F9ec8e306503d5f9c9e8fd9c0d77de8fa63428888\u002FREADME.md",[907],"Flash Next W4A8-modelkaart en evaluatiemethoden, vastgezette revisie",[32,1171,1013],{"href":1172,"ariaLabel":1173,"className":1174,"dataFootnoteBackref":37},"#user-content-fnref-fourbit","Back to reference 4",[1012],[32,1176,1013,1180],{"href":1177,"ariaLabel":1178,"className":1179,"dataFootnoteBackref":37},"#user-content-fnref-fourbit-2","Back to reference 4-2",[1012],[29,1181,47],{},[32,1183,1013,1187],{"href":1184,"ariaLabel":1185,"className":1186,"dataFootnoteBackref":37},"#user-content-fnref-fourbit-3","Back to reference 4-3",[1012],[29,1188,62],{},[998,1190,1192,1006,1197,468,1202,468,1209,468,1216,468,1223,468,1230,468,1237,468,1244],{"id":1191},"user-content-fn-release",[32,1193,1196],{"href":1194,"rel":1195},"https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin\u002Fblob\u002F4882e36e455c1e657175db910367c992837c72a5\u002Fmodels\u002FQwen3.8-Flash-Next\u002FREADME.md",[907],"Flash Next-installatie, cachemodi, hostvereisten en beperkingen van de native worker, vastgezette release",[32,1198,1013],{"href":1199,"ariaLabel":1200,"className":1201,"dataFootnoteBackref":37},"#user-content-fnref-release","Back to reference 5",[1012],[32,1203,1013,1207],{"href":1204,"ariaLabel":1205,"className":1206,"dataFootnoteBackref":37},"#user-content-fnref-release-2","Back to reference 5-2",[1012],[29,1208,47],{},[32,1210,1013,1214],{"href":1211,"ariaLabel":1212,"className":1213,"dataFootnoteBackref":37},"#user-content-fnref-release-3","Back to reference 5-3",[1012],[29,1215,62],{},[32,1217,1013,1221],{"href":1218,"ariaLabel":1219,"className":1220,"dataFootnoteBackref":37},"#user-content-fnref-release-4","Back to reference 5-4",[1012],[29,1222,213],{},[32,1224,1013,1228],{"href":1225,"ariaLabel":1226,"className":1227,"dataFootnoteBackref":37},"#user-content-fnref-release-5","Back to reference 5-5",[1012],[29,1229,230],{},[32,1231,1013,1235],{"href":1232,"ariaLabel":1233,"className":1234,"dataFootnoteBackref":37},"#user-content-fnref-release-6","Back to reference 5-6",[1012],[29,1236,725],{},[32,1238,1013,1242],{"href":1239,"ariaLabel":1240,"className":1241,"dataFootnoteBackref":37},"#user-content-fnref-release-7","Back to reference 5-7",[1012],[29,1243,1056],{},[32,1245,1013,1249],{"href":1246,"ariaLabel":1247,"className":1248,"dataFootnoteBackref":37},"#user-content-fnref-release-8","Back to reference 5-8",[1012],[29,1250,1064],{},[998,1252,1254,1006,1259],{"id":1253},"user-content-fn-launcher",[32,1255,1258],{"href":1256,"rel":1257},"https:\u002F\u002Fgithub.com\u002FEliovp-BV\u002Fpaiton-vllm-plugin\u002Fblob\u002F4882e36e455c1e657175db910367c992837c72a5\u002Fmodels\u002FQwen3.8-Flash-Next\u002Flaunch-flashnext.py",[907],"Flash Next-launcher en vastgezette downloads",[32,1260,1013],{"href":1261,"ariaLabel":1262,"className":1263,"dataFootnoteBackref":37},"#user-content-fnref-launcher","Back to reference 6",[1012],[1265,1266,1267],"style",{},"html pre.shiki code .sScJk, html code.shiki .sScJk{--shiki-default:#6F42C1;--shiki-dark:#B392F0}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":37,"searchDepth":752,"depth":752,"links":1269},[1270,1274,1275,1276,1277,1278,1279,1280],{"id":66,"depth":752,"text":67,"children":1271},[1272,1273],{"id":234,"depth":762,"text":235},{"id":304,"depth":762,"text":305},{"id":325,"depth":752,"text":326},{"id":500,"depth":752,"text":501},{"id":618,"depth":752,"text":619},{"id":656,"depth":752,"text":657},{"id":692,"depth":752,"text":693},{"id":946,"depth":752,"text":947},{"id":36,"depth":752,"text":993},[977,1282,1283,1284,1285],"AMD Radeon","Lokale AI","Qwen","Kwantisatie","2026-10-24T18:45:00Z","Paitons nieuwe 3-bit Qwen3.8-gewichten benaderen 16-bit taakscores op twee R9700-kaarten. Native MTP-prefixcaching hervat een prompt van 200K in 0,68 seconde. Resultaten, grenzen en installatie.","md","Nauwkeuriger in 3-bit. Sneller verder met je gesprek.","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen38-3bit-mtp\u002Fhero-nl.webp",{},"https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-3bit-accuracy-prefix-caching-mtp","\u002Fblog\u002Fpaiton-qwen38-3bit-accuracy-prefix-caching-mtp",{"title":5,"description":1287},"paiton-qwen38-3bit-accuracy-prefix-caching-mtp","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen38-3bit-mtp\u002Fsocial-nl.webp","blog\u002Fpaiton-qwen38-3bit-accuracy-prefix-caching-mtp",null,"YHaO3cAy71mflyYHZDNRjspOrwlI9sfNkVJ_4ZJ7SBc",[1301,1303,1318,1327,1336,1345,1355,1368,1378,1390,1400,1413,1422,1434,1465,1477,1499,1517,1536,1555,1572,1589,1601,1617,1632,1644,1653,1661,1676,1688,1699,1710,1720,1733,1743,1756,1767,1777,1788,1797,1809,1820,1829],{"path":1293,"title":5,"description":1287,"date":1286,"slug":1295,"image":1290,"originalUrl":1292,"categories":1302},[977,1282,1283,1284,1285],{"path":1304,"title":1305,"description":1306,"date":1307,"slug":1308,"image":1309,"originalUrl":1310,"categories":1311},"\u002Fblog\u002Fnvidia-dsx-pod-vera-rubin-modular-data-centers","Nvidia confirms the modular DC direction","NVIDIA DSX POD brengt Vera Rubin AI-fabrieken naar modulaire datacenters. ElioVP bespreekt terreinvoorbereiding, koeling en de aanpak achter ModFlex.","2026-10-21T13:00:00Z","nvidia-dsx-pod-vera-rubin-modular-data-centers","\u002Fasset\u002Fimages\u002Fblog\u002Fnvidia-dsx-pod-modflex\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fnvidia-dsx-pod-vera-rubin-modular-data-centers",[1312,1313,1314,1315,1316,1317],"Modular DC","AI-infrastructuur","ModFlex","NVIDIA","Vera Rubin","Vloeistofkoeling",{"path":1319,"title":1320,"description":1321,"date":1322,"slug":1323,"image":1324,"originalUrl":1325,"categories":1326},"\u002Fblog\u002Fpaiton-qwen38-flash-next-4bit-radeon-ai-pro-r9700","Qwen3.8 Flash Next in 4-bit op 2 × Radeon R9700: getest tegenover 16-bit","Paiton draait Qwen3.8 Flash Next in 4-bit op twee R9700-kaarten: 96,8% GSM8K, 96,3% HumanEval, 189,8 tokens\u002Fs en een optionele RAM-prefixcache. Kwaliteit, snelheid en installatie.","2026-10-12T15:00:00Z","paiton-qwen38-flash-next-4bit-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen38-flash-next-4bit\u002Fhero-nl.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-flash-next-4bit-radeon-ai-pro-r9700",[977,1282,1283,1284,1285],{"path":1328,"title":1329,"description":1330,"date":1331,"slug":1332,"image":1333,"originalUrl":1334,"categories":1335},"\u002Fblog\u002Fpaiton-qwen38-flash-next-radeon-ai-pro-r9700","Qwen3.8 Flash Next op 2 × Radeon AI PRO R9700: 216 tokens\u002Fs, 200K-context","Paiton draait Qwen3.8 Flash Next met 3-bit experts op twee R9700-kaarten: 216,3 tokens\u002Fs, 565,1 tokens\u002Fs samen en een aparte 200K-modus. Metingen, kwaliteit en installatie.","2026-10-10T10:30:00Z","paiton-qwen38-flash-next-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen38-flash-next\u002Fhero-nl.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-flash-next-radeon-ai-pro-r9700",[977,1282,1283,1284,1285],{"path":1337,"title":1338,"description":1339,"date":1340,"slug":1341,"image":1342,"originalUrl":1343,"categories":1344},"\u002Fblog\u002Fpaiton-qwen38-w3a4-radeon-ai-pro-r9700","Qwen3.8 27B op 1 × Radeon AI PRO R9700: 3-bit gewichten, 20% snellere decode","Qwen3.8 27B op één R9700: 19,9% snellere gewogen decode, een nieuwe 4-bit cache, 200K-context en optionele beeldinvoer. Metingen en Paiton-installatie.","2026-09-27T09:00:00Z","paiton-qwen38-w3a4-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen38-w3a4\u002Fhero-nl.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-w3a4-radeon-ai-pro-r9700",[977,1282,1283,1284,1285],{"path":1346,"title":1347,"description":1348,"date":1349,"slug":1350,"image":1351,"originalUrl":1352,"categories":1353},"\u002Fblog\u002Fpaiton-qwen-image-21-radeon-ai-pro-r9700","Qwen-Image 2.1 op 1 × Radeon AI PRO R9700: 2048×2048-beelden in 103 seconden","Genereer lokaal Qwen-Image 2.1-beelden van 2048×2048 pixels met 1 × Radeon AI PRO R9700. De Paiton v1.0.2-container mat 103,29 seconden per opgewarmd verzoek, inclusief de PNG-respons.","2026-09-23T09:00:00Z","paiton-qwen-image-21-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen-image-21\u002Fhero-nl.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen-image-21-radeon-ai-pro-r9700",[977,1282,1283,1354,1284],"Beeldgeneratie",{"path":1356,"title":1357,"description":1358,"date":1359,"slug":1360,"image":1361,"originalUrl":1362,"categories":1363},"\u002Fblog\u002Fpaiton-qwen38-mxfp4-dflash2-r9700","Qwen3.8: 400,7 tok\u002Fs op één R9700 | Paiton","Qwen3.8 op één R9700: 400,7 tok\u002Fs met ROCm 10 en vLLM 0.29, plus publieke 200K\u002F220K-chatprofielen. Benchmarks, beperkingen en startopdrachten.","2026-09-16T07:30:00Z","paiton-qwen38-mxfp4-dflash2-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-qwen38-mxfp4\u002Fupdate-2026-09-19\u002Fhero.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-mxfp4-dflash2-r9700",[977,1282,1283,1364,1365,1366,1367],"AI-inferentie","Inferentie-optimalisatie","Grote taalmodellen","vLLM",{"path":1369,"title":1370,"description":1371,"date":1372,"slug":1373,"image":1374,"originalUrl":1375,"categories":1376},"\u002Fblog\u002Fpaiton-qwen38-neo-gguf-vllm-r9700","Qwen3.8 GGUF in vLLM: sneller antwoord op één Radeon","Draai de originele NEO CODER MAX GGUF met Paiton in vLLM op een R9700. Bekijk de gemeten responstijden, beeldinvoer en lokale installatie.","2026-09-14T07:30:00Z","paiton-qwen38-neo-gguf-vllm-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-neo-gguf\u002F00-hero-neo-gguf-r9700.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-neo-gguf-vllm-r9700",[977,1282,1283,1377,1367],"GGUF",{"path":1379,"title":1380,"description":1381,"date":1382,"slug":1383,"image":1384,"originalUrl":1385,"categories":1386},"\u002Fblog\u002Fpaiton-minimax-h3-radeon-ai-pro-r9700","MiniMax H3 op Radeon: 15 seconden video met stereogeluid","Paiton genereert lokaal 15 seconden MiniMax H3-video met stereogeluid op één Radeon AI PRO R9700 in 5 min 33 s, met 16,7% minder wachttijd dan stock.","2026-09-09T07:30:00Z","paiton-minimax-h3-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-minimax-h3\u002F00-featured-minimax-h3-r9700.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-minimax-h3-radeon-ai-pro-r9700",[977,1282,1283,1387,1388,1389],"Videogeneratie","MiniMax H3","ComfyUI",{"path":1391,"title":1392,"description":1393,"date":1394,"slug":1395,"image":1396,"originalUrl":1397,"categories":1398},"\u002Fblog\u002Fpaiton-flux2-klein-radeon-ai-pro-r9700","Lokale FLUX.2 klein op Radeon AI PRO R9700: sneller beelden genereren met minder VRAM","Paiton genereert FLUX.2 klein-beelden van 1024 × 1024 in 1,054 seconden op een R9700, met 16,2% minder generatietijd en 33,4% minder piekallocatie in Torch.","2026-09-07T09:00:00","paiton-flux2-klein-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-flux2-klein\u002Ffox-paiton.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-flux2-klein-radeon-ai-pro-r9700",[977,1282,1283,1354,1399,1389],"FLUX",{"path":1401,"title":1402,"description":1403,"date":1404,"slug":1405,"image":1406,"originalUrl":1407,"categories":1408},"\u002Fblog\u002Fpaiton-ornith15-radeon-ai-pro-r9700","Ornith 1.5 haalt 44,6 tok\u002Fs op één Radeon AI PRO R9700","Paiton draait Ornith 1.5 35B A3B op één Radeon AI PRO R9700 met 44,63 outputtokens per seconde, 27% sneller en met 21,3% lagere gemodelleerde kosten.","2026-09-05T09:00:00","paiton-ornith15-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-ornith15\u002F00-featured-ornith15-r9700.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-ornith15-radeon-ai-pro-r9700",[977,1409,1282,1364,1410,1411,1365,1366,1367,1412],"Kunstmatige intelligentie","GPU-prestaties","Inferentielatentie","Kostenefficiëntie",{"path":1414,"title":1415,"description":1416,"date":1417,"slug":1418,"image":1419,"originalUrl":1420,"categories":1421},"\u002Fblog\u002Fpaiton-qwen38-radeon-ai-pro-r9700","Paiton: 21% meer Qwen3.8-doorvoer op Radeon AI PRO R9700","Paiton draait AMD's Qwen3.8 27B op één Radeon AI PRO R9700 met 39,77 outputtokens per seconde. Dat levert 21% meer throughput en 17,4% lagere gemodelleerde kosten op.","2026-09-04T09:00:00","paiton-qwen38-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-r9700\u002F00-featured-paiton-r9700.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-radeon-ai-pro-r9700",[977,1409,1282,1364,1410,1411,1365,1366,1367,1412],{"path":1423,"title":1424,"description":1425,"date":1426,"slug":1427,"image":1428,"originalUrl":1298,"categories":1429},"\u002Fblog\u002Fai-data-center-power-requirements-gpu-per-megawatt","Stroomvereisten voor AI-datacenters: de GPU-per-MW-illusie","Waarom verschillen GPU-aantallen per megawatt? Lees hoe PUE, piekbelasting, opslag, netwerken en koeling de inzetbare AI-capaciteit bepalen.","2026-07-27T23:52:00","ai-data-center-power-requirements-gpu-per-megawatt","\u002Fasset\u002Fimages\u002Fblog\u002Fai-data-center-power-requirements-gpu-per-megawatt\u002Fgpu-per-megawatt-illusion.webp",[1430,1313,1431,1314,1432,1433],"Alle","Datacenters","HPC","AMD Helios",{"path":1435,"title":1436,"description":1437,"date":1438,"slug":1439,"image":1440,"originalUrl":1441,"categories":1442},"\u002Fblog\u002Fpaiton-returns-to-its-diffusion-roots-optimizing-wan2-2-t2v-a14b-on-amd-mi355x","Wan2.2-videogeneratie: Paiton op AMD MI355X","Vergelijk Wan2.2-videogeneratie op AMD MI355X met Paiton en NVIDIA B200 via Diffusers. Lees hoe we diffusiemodellen optimaliseren.","2026-06-10T14:04:04","paiton-returns-to-its-diffusion-roots-optimizing-wan2-2-t2v-a14b-on-amd-mi355x","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaitonwan2.webp","https:\u002F\u002Feliovp.com\u002Fpaiton-returns-to-its-diffusion-roots-optimizing-wan2-2-t2v-a14b-on-amd-mi355x\u002F",[1430,1409,977,1443,1444,1445,1446,1447,1448,1449,1450,1451,1452,1453,1454,1455,1456,1315,1457,1458,977,1459,1460,1461,1462,1463,1464],"14B","AMD","B200","Benchmarks","Blackwell","Compute","Diffusie","Eliovp","Generatieve AI","GPU","Hardware","Inferentie","Instinct","MI355x","On-premises","Optimalisatie","Soevereine AI","T2V","Tekst-naar-video","Tuning","Video-generatie","Wan2.2",{"path":1466,"title":1467,"description":1468,"date":1469,"slug":1470,"image":1471,"originalUrl":1472,"categories":1473},"\u002Fblog\u002Ffrom-the-attic-to-the-front-page-eliovp-recognized-as-a-pioneer-in-chip-optimization-data-center-infrastructure","ElioVP in De Tijd: chipoptimalisatie en datacenters","Lees hoe De Tijd ElioVP belicht, van de oorsprong in chipoptimalisatie tot het werk aan modulaire datacenters en koeling voor hoge vermogensdichtheid.","2026-02-10T20:48:12","from-the-attic-to-the-front-page-eliovp-recognized-as-a-pioneer-in-chip-optimization-data-center-infrastructure","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fphysicalnewspaper.webp","https:\u002F\u002Feliovp.com\u002Ffrom-the-attic-to-the-front-page-eliovp-recognized-as-a-pioneer-in-chip-optimization-data-center-infrastructure\u002F",[1430,1409,1474,1475,1444,1476,1474,1315],"Modulaire DC","Niet gecategoriseerd","De Tijd",{"path":1478,"title":1479,"description":1480,"date":1481,"slug":1482,"image":1483,"originalUrl":1484,"categories":1485},"\u002Fblog\u002Fprivacy-is-geen-it-probleem-meer-het-is-een-strategische-prioriteit","AI en privacy: een strategische prioriteit in de Benelux","Privacyrisico's van generatieve AI, vertrouwen, dataopslag en governance. Waarom bedrijven in de Benelux veilige AI strategisch moeten benaderen.","2026-01-29T13:51:11","privacy-is-geen-it-probleem-meer-het-is-een-strategische-prioriteit","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fheaderimage.webp","https:\u002F\u002Feliovp.com\u002Fprivacy-is-geen-it-probleem-meer-het-is-een-strategische-prioriteit\u002F",[1430,1409,1486,1475,1487,1488,1489,1490,1491,1492,1493,1494,1450,1495,1451,1496,1497,1498],"Trending","AI Act","Antropomorfisme","AVG","Benelux","ChatGPT","Cyberbeveiliging","Databeheer","Gegevensbeveiliging","GDPR","Microsoft Copilot","Privacy","Shadow AI",{"path":1500,"title":1501,"description":1502,"date":1503,"slug":1504,"image":1505,"originalUrl":1506,"categories":1507},"\u002Fblog\u002Fitsme-bij-ons-is-het-its-not-me-en-dit-is-waarom","Waarom wij itsme niet gebruiken: privacy en soevereiniteit","Waarom ElioVP itsme niet gebruikt: onze afwegingen rond identiteitsmetadata, cloudafhankelijkheid, privacy en datasoevereiniteit.","2025-11-27T09:32:14","itsme-bij-ons-is-het-its-not-me-en-dit-is-waarom","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ffrontimage.webp","https:\u002F\u002Feliovp.com\u002Fitsme-bij-ons-is-het-its-not-me-en-dit-is-waarom\u002F",[1430,1508,1509,1510,1492,1511,1512,1513,1495,1514,1515,1516,1497],"AWS","Belgian Mobile ID","CLOUD Act","Datasoevereiniteit","Digitale identiteit","eIDAS","itsme","Liberty Global","MyGov.be",{"path":1518,"title":1519,"description":1520,"date":1521,"slug":1522,"image":1523,"originalUrl":1524,"categories":1525},"\u002Fblog\u002Ffield-report-the-reality-of-building-agentic-ai-in-2025","Praktijkrapport: de realiteit van Agentic AI bouwen in 2025","Praktijklessen over lokale AI-agents in 2025 gaan in op workflowontwerp, observability, modeltraining, hallucinaties en GPU-geheugenlimieten.","2025-11-25T14:03:39","field-report-the-reality-of-building-agentic-ai-in-2025","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ffieldreport.webp","https:\u002F\u002Feliovp.com\u002Ffield-report-the-reality-of-building-agentic-ai-in-2025\u002F",[1430,1409,1526,1486,1527,1528,1529,1530,1531,1532,1533,1534,1459,1535],"Oplossingen","Agentic AI","AI-techniek","AI-strategie","Autonome agenten","Bedrijfs-AI","Lokale LLM","Modelverfijning","AI op locatie","VRAM-optimalisatie",{"path":1537,"title":1538,"description":1539,"date":1540,"slug":1541,"image":1542,"originalUrl":1543,"categories":1544},"\u002Fblog\u002Fthe-synthetic-unicorn-bubble","De synthetische unicornzeepbel","Een analyse van investeringsrisico’s bij AI-neoclouds: circulaire financiering, infrastructuurclaims, contractvoorwaarden en due diligence.","2025-11-24T19:22:28","the-synthetic-unicorn-bubble","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fsyntheticunicorn.webp","https:\u002F\u002Feliovp.com\u002Fthe-synthetic-unicorn-bubble\u002F",[1430,1409,1486,1545,1546,1547,1548,1549,1550,1551,1552,1553,1554],"AI Infrastructure","AI Neocloud","Circulaire financiering","GPU Cloud","Beleggingsrisico's","Opstartwaardering","Synthetische bubbel","Technische analyse","Vaporware","Durfkapitaal",{"path":1556,"title":1557,"description":1558,"date":1559,"slug":1560,"image":1561,"originalUrl":1562,"categories":1563},"\u002Fblog\u002Fbuilding-the-engine-for-the-ai-race-the-4-month-path-to-nvidia-gb300-nvl72-power","NVIDIA GB300 NVL72: modulair datacenter in vier maanden","Ontdek een modulair datacenterontwerp voor NVIDIA GB300 NVL72, met redundante voeding, hybride koeling en een bouwplanning van vier maanden.","2025-11-20T14:10:19","building-the-engine-for-the-ai-race-the-4-month-path-to-nvidia-gb300-nvl72-power","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fsuperpodmodflexfrontimage.webp","https:\u002F\u002Feliovp.com\u002Fbuilding-the-engine-for-the-ai-race-the-4-month-path-to-nvidia-gb300-nvl72-power\u002F",[1430,1474,1475,1564,1313,1565,1566,1317,1567,1568,1569,1570,1571],"150 kW-rack","DLC","Hoge dichtheid","Modulair datacenter","NVIDIA Blackwell Ultra","NVIDIA GB300","NVL72","Snelle implementatie",{"path":1573,"title":1574,"description":1575,"date":1576,"slug":1577,"image":1578,"originalUrl":1579,"categories":1580},"\u002Fblog\u002Fwhy-cuda-translation-wont-unlock-amds-real-potential","CUDA-vertaling versus AMD-gerichte optimalisatie","Waarom CUDA-compatibiliteit niet hetzelfde is als AMD-prestaties: over ROCm, HIP, kerneloptimalisatie en hardwaregerichte afstemming.","2025-11-12T14:48:37","why-cuda-translation-wont-unlock-amds-real-potential","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fchatgpt-image-nov-11-2025-09_16_10-pm-1.webp","https:\u002F\u002Feliovp.com\u002Fwhy-cuda-translation-wont-unlock-amds-real-potential\u002F",[1430,1409,977,1475,1581,1409,1582,1583,1584,1585,1586,1587,977,1588],"AMD MI300X","CUDA-vertaling","FP8","GPU-optimalisatie","High-performance computing","HIP","Kerneltuning","ROCm",{"path":1590,"title":1591,"description":1592,"date":1593,"slug":1594,"image":1595,"originalUrl":1596,"categories":1597},"\u002Fblog\u002Fpaiton-the-simplest-way-to-supercharge-ai-inference","Paiton: snellere AI-inferentie in uw bestaande stack","Lees hoe Paiton aansluit op bestaande inferentiestacks, met AMD MI300X-benchmarks en vergelijkingen van prestaties per dollar.","2025-11-11T10:31:22","paiton-the-simplest-way-to-supercharge-ai-inference","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaiton-powaaah.webp","https:\u002F\u002Feliovp.com\u002Fpaiton-the-simplest-way-to-supercharge-ai-inference\u002F",[1430,1409,977,1364,1598,1581,1412,1599,1365,1587,977,1600,1367],"AMD Instinct","Hoge throughput","SGLang",{"path":1602,"title":1603,"description":1604,"date":1605,"slug":1606,"image":1607,"originalUrl":1608,"categories":1609},"\u002Fblog\u002Fstop-overpaying-paiton-mi300x-moe-beats-h200-b200-on-1m-tokens","Paiton MoE-benchmarks: MI300X versus H200 en B200","Vergelijk Qwen3-30B-A3B MoE-benchmarks van MI300X met Paiton, H200 en B200: throughput en kosten per miljoen tokens.","2025-09-26T13:36:18","stop-overpaying-paiton-mi300x-moe-beats-h200-b200-on-1m-tokens","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fhulkvshulkpaitonwins.webp","https:\u002F\u002Feliovp.com\u002Fstop-overpaying-paiton-mi300x-moe-beats-h200-b200-on-1m-tokens\u002F",[1430,1409,977,1610,1581,1611,1365,1612,1613,1614,1615,977,1616],"AI-benchmarks","Kosten per token","Mixture of Experts","MoE","NVIDIA B200","NVIDIA H200","Qwen3",{"path":1618,"title":1619,"description":1620,"date":1621,"slug":1622,"image":1623,"originalUrl":1624,"categories":1625},"\u002Fblog\u002Fagentic-ai-but-make-it-local-from-inbox-to-insight-to-action-en","Lokale Agentic AI: van inbox naar actie","Lokale AI-agents zetten e-mails, documenten en beelden om in tickets, rapporten en acties, met modellen op maat van uw gegevens en systemen.","2025-09-16T13:09:00","agentic-ai-but-make-it-local-from-inbox-to-insight-to-action-en","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ffrontfotoblog.webp","https:\u002F\u002Feliovp.com\u002Fagentic-ai-but-make-it-local-from-inbox-to-insight-to-action-en\u002F",[1430,1409,1526,1475,1527,1626,1627,1628,1629,1532,1534,1459,1630,1631],"Schadedetectie","Documentverwerking","E-mailautomatisering","Factuurextractie","Ticketautomatisering","Workflowautomatisering",{"path":1633,"title":1634,"description":1635,"date":1636,"slug":1637,"image":1638,"originalUrl":1639,"categories":1640},"\u002Fblog\u002Fmi300x-fp8-data-parallel-benchmarks-8-64-gpus-h200-left-behind-b200-within-reach","MI300X FP8-benchmarks: GPU-partitionering met Paiton","Bekijk hoe Paiton presteert met Llama 3.1 8B FP8 op gepartitioneerde MI300X-GPU's, vergeleken met NVIDIA H200 en B200.","2025-07-31T13:32:57","mi300x-fp8-data-parallel-benchmarks-8-64-gpus-h200-left-behind-b200-within-reach","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fimage-2-1.webp","https:\u002F\u002Feliovp.com\u002Fmi300x-fp8-data%e2%80%91parallel-benchmarks-8-64-gpus-h200-left-behind-b200-within-reach\u002F",[1430,1409,977,1475,1641,1444,1445,1642,1643,1456,1315,977,1367],"AI","H200","MI300X",{"path":1645,"title":1646,"description":1647,"date":1648,"slug":1649,"image":1650,"originalUrl":1651,"categories":1652},"\u002Fblog\u002Fapplicable-ai-for-businesses","Toepasbare AI voor bedrijven","Ontdek hoe ElioVP lokale AI voor bedrijfsprocessen bouwt, met modeltraining op maat en automatische schadedetectie voor de logistiek.","2025-07-09T21:35:30","applicable-ai-for-businesses","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fscherm_afbeelding-2025-07-09-om-23.30.17.webp","https:\u002F\u002Feliovp.com\u002Fapplicable-ai-for-businesses\u002F",[1430,1409,1526,1527,1626,1627,1628,1629,1532,1534,1459,1630,1631],{"path":1654,"title":1655,"description":1656,"date":1657,"slug":1658,"image":37,"originalUrl":1659,"categories":1660},"\u002Fblog\u002Fintroducing-paitons-free-evaluation-models","Maak kennis met de gratis evaluatiemodellen van Paiton","Test Paiton met gratis evaluatiemodellen voor AMD-GPU's. Vergelijk de prestaties voor tekst, beeldanalyse en beeldgeneratie met uw eigen workloads.","2025-07-07T11:26:13","introducing-paitons-free-evaluation-models","https:\u002F\u002Feliovp.com\u002Fintroducing-paitons-free-evaluation-models\u002F",[1430,1409,977],{"path":1662,"title":1663,"description":1664,"date":1665,"slug":1666,"image":1667,"originalUrl":1668,"categories":1669},"\u002Fblog\u002Fpaiton-dramatically-faster-startup-and-performance-for-llama-3-1-405b","Llama 3.1 405B: sneller starten met Paiton op MI300X","Bekijk Paiton-benchmarks voor Llama 3.1 405B op acht AMD MI300X-GPU's, met opstarttijd, tensorparallelisme, throughput en latency.","2025-06-12T20:15:23","paiton-dramatically-faster-startup-and-performance-for-llama-3-1-405b","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fservingscreenshot.webp","https:\u002F\u002Feliovp.com\u002Fpaiton-dramatically-faster-startup-and-performance-for-llama-3-1-405b\u002F",[1430,1409,977,1475,1364,1581,1670,1583,1671,1672,1673,977,1674,1675],"Koude start","Grafiekcompilatie","Llama 3.1 405B","LLM-optimalisatie","Opstartlatentie","Tensor-parallellisme",{"path":1677,"title":1678,"description":1679,"date":1680,"slug":1681,"image":1682,"originalUrl":1683,"categories":1684},"\u002Fblog\u002Fpaiton-fp8-beats-nvidias-h200-on-amds-mi300x","Paiton FP8 verslaat NVIDIA's H200 op AMD's MI300X","Vergelijk Paiton op AMD MI300X met NVIDIA H200 voor Llama 3.1 70B FP8: throughput, wachttijd tot het eerste token en latency bij diverse batchgroottes.","2025-06-08T19:12:40","paiton-fp8-beats-nvidias-h200-on-amds-mi300x","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fblognewfp8.webp","https:\u002F\u002Feliovp.com\u002Fpaiton-fp8-beats-nvidias-h200-on-amds-mi300x\u002F",[1430,1409,977,1475,1581,1685,1531,1451,1410,1411,1366,1672,1686,1687],"Koude startoptimalisatie","Model serving","vLLM-optimalisatie",{"path":1689,"title":1690,"description":1691,"date":1692,"slug":1693,"image":1694,"originalUrl":1695,"categories":1696},"\u002Fblog\u002Fmi300x-vs-h200-vs-rx-7900-xtx-vs-tenstorrent-n300s-with-vllm","MI300X, H200, RX 7900 XTX en n300s: vLLM-benchmarks","Vergelijk MI300X, H200, RX 7900 XTX en Tenstorrent n300s met vLLM: throughput, gemodelleerde tokenkosten en hardwarebeperkingen voor Llama 3 8B.","2025-05-09T14:03:58","mi300x-vs-h200-vs-rx-7900-xtx-vs-tenstorrent-n300s-with-vllm","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fcomparisontenstor.webp","https:\u002F\u002Feliovp.com\u002Fmi300x-vs-h200-vs-rx-7900-xtx-vs-tenstorrent-n300s-with-vllm\u002F",[1430,1409,977,1526,1475,1444,1643,1315,1697,1698],"RX7900XTX","tenstorrent",{"path":1700,"title":1701,"description":1702,"date":1703,"slug":1704,"image":1705,"originalUrl":1706,"categories":1707},"\u002Fblog\u002Fclusterpl-empowering-gpu-cluster-investors-with-real-world-financial-insights","ClusterP&L: financiële modellen voor GPU-clusters","Ontdek hoe ClusterP&L kosten, rendement en investeringsscenario's voor GPU-clusters modelleert, met risicoanalyses en exporteerbare rapporten.","2025-05-03T10:52:22","clusterpl-empowering-gpu-cluster-investors-with-real-world-financial-insights","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fcomparisonscenarios.webp","https:\u002F\u002Feliovp.com\u002Fclusterpl-empowering-gpu-cluster-investors-with-real-world-financial-insights\u002F",[1430,1409,1474,1526,1445,1642,1708,1315,1709],"MI325X","P&L-calculator",{"path":1711,"title":1712,"description":1713,"date":1714,"slug":1715,"image":1716,"originalUrl":1717,"categories":1718},"\u002Fblog\u002Fcranking-out-faster-tokens-for-fewer-dollars-amd-mi300x-vs-nvidia-h200","AMD MI300X versus NVIDIA H200: Qwen3-32B met Paiton","Vergelijk Qwen3-32B-benchmarks op AMD MI300X met Paiton en NVIDIA H200, met resultaten voor throughput, latency en hardwarekosten.","2025-05-02T21:10:30","cranking-out-faster-tokens-for-fewer-dollars-amd-mi300x-vs-nvidia-h200","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002F3ac59a73-2466-4422-b7e5-ef2e4a8ca58e.webp","https:\u002F\u002Feliovp.com\u002Fcranking-out-faster-tokens-for-fewer-dollars-amd-mi300x-vs-nvidia-h200\u002F",[1430,1409,977,1641,1444,1642,1719,1315,977,1367],"MI300",{"path":1721,"title":1722,"description":1723,"date":1724,"slug":1725,"image":1726,"originalUrl":1727,"categories":1728},"\u002Fblog\u002Fpower-meets-precision-high-density-modular-data-center-for-nvidia-nvl-deployments-1-2-mw","Modulaire datacenters voor NVIDIA NVL: 1 tot 2 MW","Ontdek modulaire datacenterontwerpen voor NVIDIA NVL-systemen, met aandacht voor vermogen, vloeistofkoeling, redundantie en uitrolplanning.","2025-05-02T14:09:59","power-meets-precision-high-density-modular-data-center-for-nvidia-nvl-deployments-1-2-mw","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Feliovp_critical-1mw-pod_rev-2_transparent.webp","https:\u002F\u002Feliovp.com\u002Fpower-meets-precision-high-density-modular-data-center-for-nvidia-nvl-deployments-1-2-mw\u002F",[1430,1474,1729,1545,1566,1432,1317,1567,1730,1731,1570,1732],"1-2MW datacenter","NVIDIA Blackwell","NVIDIA NVL","Precisiekoeling",{"path":1734,"title":1735,"description":1736,"date":1737,"slug":1738,"image":1739,"originalUrl":1740,"categories":1741},"\u002Fblog\u002Fexamining-ai-agents-in-the-medical-field-ai-that-speaks-dicom","AI-agents in de medische wereld: AI die DICOM spreekt","Ontdek een lokale AI-agent die DICOM-gegevens opzoekt en bekijk tests van beeldmodellen met geanonimiseerde medische beelden.","2025-04-11T14:45:48","examining-ai-agents-in-the-medical-field-ai-that-speaks-dicom","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fhealthcareblog-1.webp","https:\u002F\u002Feliovp.com\u002Fexamining-ai-agents-in-the-medical-field-ai-that-speaks-dicom\u002F",[1430,1409,1526,1475,1641,1444,1742],"Zorg",{"path":1744,"title":1745,"description":1746,"date":1747,"slug":1748,"image":1749,"originalUrl":1750,"categories":1751},"\u002Fblog\u002Feliovp-bv-your-trusted-partner-for-supply-chain-resilience-amidst-new-u-s-tariffs","Amerikaanse invoerheffingen en AI-leveringszekerheid in 2025","Lees ElioVP's visie uit april 2025 op Amerikaanse invoerheffingen en leveringszekerheid voor AI-servers, HPC-systemen en modulaire datacenters.","2025-04-04T10:01:27","eliovp-bv-your-trusted-partner-for-supply-chain-resilience-amidst-new-u-s-tariffs","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ftariffsshipping.webp","https:\u002F\u002Feliovp.com\u002Feliovp-bv-your-trusted-partner-for-supply-chain-resilience-amidst-new-u-s-tariffs\u002F",[1430,1486,1641,1444,1752,1753,1754,1755],"Invoer","Taiwan","Invoerheffingen","Trump",{"path":1757,"title":1758,"description":1759,"date":1760,"slug":1761,"image":1762,"originalUrl":1763,"categories":1764},"\u002Fblog\u002Fwhy-ai-agents-are-the-future","Waarom AI-agenten de toekomst zijn","Ontdek AI-agents voor ERP, CRM, financiën en klantondersteuning, met praktijkvoorbeelden en een traject van procesanalyse tot pilot en uitrol.","2025-03-23T22:06:59","why-ai-agents-are-the-future","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ferp2.jpeg","https:\u002F\u002Feliovp.com\u002Fwhy-ai-agents-are-the-future\u002F",[1430,1409,1526,1641,1765,1766],"AI-agenten","ERP",{"path":1768,"title":1769,"description":1770,"date":1771,"slug":1772,"image":1773,"originalUrl":1774,"categories":1775},"\u002Fblog\u002Fthe-rise-of-open-source-ai-model-optimization","De opkomst van open-source AI-modeloptimalisatie","Verken trends in opensource-AI-optimalisatie: kwantisatie, mixture-of-experts-modellen, hardwaregerichte afstemming, RAG en edge-AI.","2025-03-22T20:59:23","the-rise-of-open-source-ai-model-optimization","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Friseofopensource.jpeg","https:\u002F\u002Feliovp.com\u002Fthe-rise-of-open-source-ai-model-optimization\u002F",[1430,1409,1486,1776,1444,1452,1315],"AI-nieuws",{"path":1778,"title":1779,"description":1780,"date":1781,"slug":1782,"image":1783,"originalUrl":1784,"categories":1785},"\u002Fblog\u002Fintroducing-our-benchmarking-tool-powered-by-dstack","Maak kennis met onze benchmarktool, gebouwd op dstack","Ontdek onze benchmarktool met dstack: herhaalbare vLLM-tests, automatische parameterreeksen en prestatierapporten voor lokale GPU's en de cloud.","2025-03-20T14:21:59","introducing-our-benchmarking-tool-powered-by-dstack","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fbenchmarktool.jpeg","https:\u002F\u002Feliovp.com\u002Fintroducing-our-benchmarking-tool-powered-by-dstack\u002F",[1430,1409,977,1641,1444,1786,1787,1643,977],"benchmark","LLM",{"path":1789,"title":1790,"description":1791,"date":1792,"slug":1793,"image":1794,"originalUrl":1795,"categories":1796},"\u002Fblog\u002Foptimizing-qwq-32b-by-qwen-amd-mi300x-vs-nvidia-h200","QwQ-32B optimaliseren (door Qwen): AMD MI300X versus NVIDIA H200","Vergelijk throughput en latency van QwQ-32B op AMD MI300X met Paiton en NVIDIA H200, bij kleine batches en meer gelijktijdige aanvragen.","2025-03-19T21:41:44","optimizing-qwq-32b-by-qwen-amd-mi300x-vs-nvidia-h200","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaiton4.jpeg","https:\u002F\u002Feliovp.com\u002Foptimizing-qwq-32b-by-qwen-amd-mi300x-vs-nvidia-h200\u002F",[1430,1409,977],{"path":1798,"title":1799,"description":1800,"date":1801,"slug":1802,"image":1803,"originalUrl":1804,"categories":1805},"\u002Fblog\u002Feliovp-featured-on-amd-tech-talk-podcast","Eliovp te gast in de AMD Tech Talk-podcast","Beluister Elio Van Puyvelde en Jim Greene in de AMD Tech Talk-podcast over het ontstaan van ElioVP en de hardware- en softwarediensten voor AI.","2025-03-19T07:53:39","eliovp-featured-on-amd-tech-talk-podcast","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ftechtalkjimgreene.jpeg","https:\u002F\u002Feliovp.com\u002Feliovp-featured-on-amd-tech-talk-podcast\u002F",[1430,1444,1806,1807,1808],"Jim Greene","Podcast","Tech Talk",{"path":1810,"title":1811,"description":1812,"date":1813,"slug":1814,"image":1815,"originalUrl":1816,"categories":1817},"\u002Fblog\u002Ffurther-optimizing-amd-powered-inference-with-paiton","AMD-inferentie verder optimaliseren met Paiton","Bekijk Paiton-benchmarks voor DeepSeek R1 Distill Llama 8B op AMD MI300X, gericht op throughput en latency bij kleinere batchgroottes.","2025-03-13T06:18:30","further-optimizing-amd-powered-inference-with-paiton","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaitonpost3.webp","https:\u002F\u002Feliovp.com\u002Ffurther-optimizing-amd-powered-inference-with-paiton\u002F",[1430,1409,977,1444,1818,1819,1642,1643,1708,977,1367],"DeepSeek","H100",{"path":1821,"title":1822,"description":1823,"date":1824,"slug":1825,"image":1826,"originalUrl":1827,"categories":1828},"\u002Fblog\u002Fa-first-look-at-paiton-in-action-deepseek-r1-distill-llama-3-1-8b","Paiton-benchmarks: DeepSeek R1 Distill Llama 3.1 8B","Vergelijk standaard- en Paiton-versies van DeepSeek R1 Distill Llama 3.1 8B op AMD MI300X, met benchmarks voor throughput en latency.","2025-01-31T09:11:02","a-first-look-at-paiton-in-action-deepseek-r1-distill-llama-3-1-8b","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaitonpost2.webp","https:\u002F\u002Feliovp.com\u002Fa-first-look-at-paiton-in-action-deepseek-r1-distill-llama-3-1-8b\u002F",[1430,1409,977,1444,1818,1819,1642,1643,1708,977,1367],{"path":1830,"title":1831,"description":1832,"date":1833,"slug":1834,"image":1835,"originalUrl":1836,"categories":1837},"\u002Fblog\u002Fai-model-optimization-with-paiton","AI-modeloptimalisatie met Paiton","Lees hoe Paiton modelcompilatie, aangepaste kernels en kernelfusie inzet om AI-inferentie op AMD GPU's te optimaliseren.","2025-01-30T19:53:25","ai-model-optimization-with-paiton","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaitonpost1.webp","https:\u002F\u002Feliovp.com\u002Fai-model-optimization-with-paiton\u002F",[1430,1409,977,1444,1819,1642,1643,1708,977,1367],1792870334450]