{"id":6388,"date":"2025-05-08T03:35:21","date_gmt":"2025-05-08T03:35:21","guid":{"rendered":"https:\/\/www.myshirtai.com\/archives\/6388"},"modified":"2025-05-08T03:35:21","modified_gmt":"2025-05-08T03:35:21","slug":"nvidia-llama-nemotron%ef%bc%9a%e8%b6%85%e8%b6%8adeepseek-r1%e7%9a%84%e5%bc%80%e6%ba%90%e6%96%b0%e7%8e%8b%e8%80%85","status":"publish","type":"post","link":"https:\/\/www.myshirtai.com\/pt\/archives\/6388","title":{"rendered":"NVIDIA Llama-Nemotron: O novo rei do c\u00f3digo aberto para al\u00e9m do DeepSeek-R1"},"content":{"rendered":"<h2 class=\"wp-block-heading\" id=\"h-\u82f1\u4f1f\u8fbe\u5f00\u6e90\u65b0\u9738\u4e3b-\u4ece6710\u4ebf\u52302530\u4ebf\u53c2\u6570\u7684\u6548\u7387\u9769\u547d\">O novo juggernaut de c\u00f3digo aberto da NVIDIA: uma revolu\u00e7\u00e3o na efici\u00eancia de 671 mil milh\u00f5es para 253 mil milh\u00f5es de par\u00e2metros<\/h2>\n\n\n\n<p>Na era atual de r\u00e1pido desenvolvimento de grandes modelos de IA, a NVIDIA est\u00e1 mais uma vez a fazer ondas com as suas proezas tecnol\u00f3gicas. Recentemente, a NVIDIA lan\u00e7ou a s\u00e9rie de modelos Llama-Nemotron, que rapidamente ascendeu ao topo dos modelos de c\u00f3digo aberto com uma efici\u00eancia e desempenho surpreendentes, ultrapassando mesmo o DeepSeek-R1, que tem um n\u00famero muito maior de par\u00e2metros, numa s\u00e9rie de refer\u00eancias importantes.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" src=\"https:\/\/school.myshirtai.com\/wp-content\/uploads\/2025\/05\/image-16.png\" alt=\"\" class=\"wp-image-1014\"\/><\/figure>\n\n\n\n<p>A s\u00e9rie Llama-Nemotron cont\u00e9m tr\u00eas modelos:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>LN-Nano (8B)<\/strong>Miniaturas eficientes concebidas para dispositivos de ponta e aplica\u00e7\u00f5es m\u00f3veis<\/li>\n\n\n\n<li><strong>LN-Super (49B)<\/strong>Um modelo de gama m\u00e9dia que equilibra desempenho e efici\u00eancia<\/li>\n\n\n\n<li><strong>LN-Ultra (253B)<\/strong>Modelo de infer\u00eancia emblem\u00e1tico concebido para tarefas complexas<\/li>\n<\/ul>\n\n\n\n<p>O mais surpreendente \u00e9 que o LN-Ultra supera o DeepSeek-R1 em v\u00e1rios benchmarks importantes, como GPQA-Diamond (76,01 vs. 71,5), IFEval (89,45 vs. 83,3) e LiveCodeBench (66,31), com apenas 253 mil milh\u00f5es de par\u00e2metros (cerca de um ter\u00e7o dos 671 mil milh\u00f5es de par\u00e2metros do DeepSeek-R1). Em benchmarks, incluindo GPQA-Diamond (76,01 vs. 71,31), IFEval (8,45 vs. 71,45) e LiveCodeBench (66,31), o LN-Ultra supera o DeepSeek-R1 em todos os aspectos e, mais importante, o LN-Ultra \u00e9 executado de forma eficiente em um \u00fanico n\u00f3 8xH100, enquanto o DeepSeek-R1 requer hardware 8xH200, o que significa que n\u00e3o s\u00f3 tem melhor desempenho, mas tamb\u00e9m oferece maior rendimento no racioc\u00ednio e um limite mais baixo para implanta\u00e7\u00e3o.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" src=\"https:\/\/school.myshirtai.com\/wp-content\/uploads\/2025\/05\/image-18.png\" alt=\"\" class=\"wp-image-1016\"\/><\/figure>\n\n\n\n<p>De acordo com o \u00cdndice de Intelig\u00eancia Anal\u00edtica Artificial, a partir de abril de 2025, o Llama-Nemotron-Ultra foi reconhecido como o modelo de c\u00f3digo aberto \"mais inteligente\" dispon\u00edvel. Esta s\u00e9rie de modelos, todos ao abrigo de licen\u00e7as de fonte aberta favor\u00e1veis \u00e0s empresas, a Licen\u00e7a de Modelo Aberto NVIDIA e a Licen\u00e7a Comunit\u00e1ria Llama, permite que as empresas os utilizem e modifiquem livremente, o que ir\u00e1, sem d\u00favida, acelerar a popularidade da tecnologia de IA e a inova\u00e7\u00e3o das aplica\u00e7\u00f5es.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h-\u6a21\u578b\u8bad\u7ec3\u63ed\u79d8-14\u4e07h100\u5c0f\u65f6\u7684\u4e94\u9636\u6bb5\u6784\u5efa\u6d41\u7a0b\">Forma\u00e7\u00e3o de modelos revelada: um processo de constru\u00e7\u00e3o em cinco fases para 140 000 horas H100<\/h2>\n\n\n\n<p>A NVIDIA revelou o processo de constru\u00e7\u00e3o em cinco fases da fam\u00edlia de modelos Llama-Nemotron num relat\u00f3rio t\u00e9cnico, mostrando todos os pormenores t\u00e9cnicos, desde a otimiza\u00e7\u00e3o da arquitetura \u00e0 aprendizagem por refor\u00e7o.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-\u7b2c\u4e00\u9636\u6bb5-\u795e\u7ecf\u67b6\u6784\u641c\u7d22\u4e0effn\u878d\u5408\">Fase 1: Pesquisa de arquitetura neural com fus\u00e3o FFN<\/h3>\n\n\n\n<p>A equipa come\u00e7ou por otimizar profundamente a arquitetura original baseada no Llama 3.1 utilizando uma estrutura de Pesquisa de Arquitetura Neural (NAS) chamada Puzzle. As varia\u00e7\u00f5es foram implementadas atrav\u00e9s da constru\u00e7\u00e3o de uma biblioteca de m\u00f3dulos transformadores alternativos:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Mecanismo de aten\u00e7\u00e3o seletivamente removido para reduzir a computa\u00e7\u00e3o e o consumo de mem\u00f3ria cache KV<\/li>\n\n\n\n<li>Dimens\u00f5es vari\u00e1veis de FFN para compress\u00e3o de modelos em diferentes granularidades<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" src=\"https:\/\/school.myshirtai.com\/wp-content\/uploads\/2025\/05\/image-17.png\" alt=\"\" class=\"wp-image-1015\"\/><\/figure>\n\n\n\n<p>Particularmente inovadora \u00e9 a tecnologia FFN Fusion (FFN Fusion): quando blocos FFN cont\u00ednuos aparecem no modelo depois de o NAS remover algumas das camadas de aten\u00e7\u00e3o, a FFN Fusion substitui estas estruturas por menos camadas FFN execut\u00e1veis em paralelo, mas mais largas, o que melhora significativamente a efici\u00eancia computacional num ambiente multi-GPU.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-\u7b2c\u4e8c\u9636\u6bb5-\u77e5\u8bc6\u84b8\u998f\u4e0e\u6301\u7eed\u9884\u8bad\u7ec3\">Fase 2: Destila\u00e7\u00e3o de conhecimentos e pr\u00e9-forma\u00e7\u00e3o cont\u00ednua<\/h3>\n\n\n\n<p>Ap\u00f3s a otimiza\u00e7\u00e3o da arquitetura, a equipa realizou uma destila\u00e7\u00e3o de conhecimentos em grande escala com pr\u00e9-treino cont\u00ednuo para recuperar e melhorar o desempenho do modelo:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>O LN-Super treina 40 mil milh\u00f5es de tokens utilizando o conjunto de dados Distillation Mix<\/li>\n\n\n\n<li>O LN-Ultra come\u00e7a por treinar o mesmo conjunto de dados para 65 mil milh\u00f5es de tokens e depois continua a treinar 88 mil milh\u00f5es de tokens no conjunto de dados da fase 4 do Nemotron-H<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" src=\"https:\/\/school.myshirtai.com\/wp-content\/uploads\/2025\/05\/image-21.png\" alt=\"\" class=\"wp-image-1020\"\/><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-\u7b2c\u4e09\u9636\u6bb5-\u5408\u6210\u6570\u636e\u76d1\u7763\u5fae\u8c03\">Fase III: S\u00edntese de dados para monitorizar a afina\u00e7\u00e3o<\/h3>\n\n\n\n<p>A fase de afina\u00e7\u00e3o supervisionada utiliza uma metodologia inovadora de forma\u00e7\u00e3o de dados sint\u00e9ticos que constr\u00f3i cuidadosamente conjuntos de dados contendo amostras inferenciais e n\u00e3o inferenciais:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Exemplo de racioc\u00ednio: \"Pensamento pormenorizado sobre\" adicionado ao comando do sistema.<\/li>\n\n\n\n<li>Amostras sem racioc\u00ednio: utiliza\u00e7\u00e3o do \"pensamento pormenorizado\"<\/li>\n<\/ul>\n\n\n\n<p>Esta conce\u00e7\u00e3o permite que o modelo alterne dinamicamente os comportamentos de infer\u00eancia de acordo com o conte\u00fado da pista, lan\u00e7ando as bases para a fun\u00e7\u00e3o de \"troca de infer\u00eancia\".<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-\u7b2c\u56db\u9636\u6bb5-\u5927\u89c4\u6a21\u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3\">Fase IV: Forma\u00e7\u00e3o intensiva e maci\u00e7a em mat\u00e9ria de aprendizagem<\/h3>\n\n\n\n<p>Esta fase \u00e9 fundamental para que o LN-Ultra ultrapasse o DeepSeek-R1. A equipa utilizou o mesmo algoritmo Grouped Relative Policy Optimisation (GRPO) que o DeepSeek-R1, e a conce\u00e7\u00e3o inovadora do processo de forma\u00e7\u00e3o incluiu:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Incentivos: incentivos \u00e0 exatid\u00e3o (baseados na correspond\u00eancia de respostas padr\u00e3o) e incentivos ao formato (para for\u00e7ar a utiliza\u00e7\u00e3o de etiquetas espec\u00edficas)<\/li>\n\n\n\n<li>Triagem de dados: as amostras simples com uma taxa de aprova\u00e7\u00e3o \u226575% foram pr\u00e9-censuradas<\/li>\n\n\n\n<li>Forma\u00e7\u00e3o em curso: atribui\u00e7\u00e3o progressiva de lotes com base na taxa de aprova\u00e7\u00e3o, com transi\u00e7\u00e3o gradual de amostras f\u00e1ceis para amostras dif\u00edceis<\/li>\n<\/ul>\n\n\n\n<p>Todo o processo de forma\u00e7\u00e3o consome cerca de 140.000 horas de GPU H100, utiliza 72 n\u00f3s (8 GPUs H100 por n\u00f3) e emprega a precis\u00e3o FP8 na fase de gera\u00e7\u00e3o e a precis\u00e3o BF16 na fase de forma\u00e7\u00e3o, que \u00e9 uma combina\u00e7\u00e3o de t\u00e9cnicas que permite ao LN-Ultra obter melhorias significativas de precis\u00e3o no conjunto de dados GPQA-Diamond.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" src=\"https:\/\/school.myshirtai.com\/wp-content\/uploads\/2025\/05\/image-19.png\" alt=\"\" class=\"wp-image-1017\"\/><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"h-\u7b2c\u4e94\u9636\u6bb5-\u6307\u4ee4\u5bf9\u9f50\u4e0e\u4eba\u7c7b\u504f\u597d\u4f18\u5316\">Fase 5: Alinhamento de comandos e otimiza\u00e7\u00e3o das prefer\u00eancias humanas<\/h3>\n\n\n\n<p>Na fase final, foi realizada uma breve sess\u00e3o de aprendizagem por refor\u00e7o, centrada na otimiza\u00e7\u00e3o das capacidades de seguimento de comandos do modelo e no alinhamento das prefer\u00eancias humanas. A equipa utilizou a tecnologia RLHF para melhorar a capacidade de ajuda geral do modelo e o desempenho do chat, mantendo a sua capacidade em \u00e1reas especializadas como a matem\u00e1tica e as ci\u00eancias. Os resultados mostraram que o LN-Super alinhado obteve 88,3 pontos no teste Arena Hard, superando modelos propriet\u00e1rios como o Claude 3.5 Sonnet e o GPT-4o.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" src=\"https:\/\/school.myshirtai.com\/wp-content\/uploads\/2025\/05\/image-22.png\" alt=\"\" class=\"wp-image-1021\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h-\u9769\u547d\u6027\u521b\u65b0-\u63a8\u7406\u5f00\u5173\u529f\u80fd\u4e0e\u786c\u4ef6\u611f\u77e5\u4f18\u5316\">Inova\u00e7\u00e3o revolucion\u00e1ria: Funcionalidade de comuta\u00e7\u00e3o de infer\u00eancias e otimiza\u00e7\u00e3o da sensibiliza\u00e7\u00e3o do hardware<\/h2>\n\n\n\n<p>Uma das maiores inova\u00e7\u00f5es da s\u00e9rie Llama-Nemotron \u00e9 a fun\u00e7\u00e3o de comuta\u00e7\u00e3o de racioc\u00ednio, que permite ao utilizador alternar dinamicamente entre os dois modos, bastando para isso acrescentar \"pensamento detalhado ligado\/desligado\" ao prompt do sistema:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Modo de conversa\u00e7\u00e3o padr\u00e3o<\/strong>Responder rapidamente aos pedidos de informa\u00e7\u00e3o di\u00e1rios com respostas diretas<\/li>\n\n\n\n<li><strong>modelo de infer\u00eancia profunda<\/strong>Racioc\u00ednio complexo em v\u00e1rias etapas, demonstrando um processo de pensamento completo<\/li>\n<\/ul>\n\n\n\n<p>Esta conce\u00e7\u00e3o resolve um dos principais problemas dos actuais modelos de IA - os programadores n\u00e3o precisam de manter modelos com diferentes arquitecturas e podem ajustar de forma flex\u00edvel os comportamentos dos modelos de acordo com a procura. No espa\u00e7o global de c\u00f3digo aberto da IA, esta \u00e9 a primeira fam\u00edlia de modelos a implementar esta funcionalidade.<\/p>\n\n\n\n<p>Ao n\u00edvel da otimiza\u00e7\u00e3o do hardware, a s\u00e9rie Nemotron foi submetida a uma profunda otimiza\u00e7\u00e3o consciente do hardware:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Apoio \u00e0 precis\u00e3o<\/strong>BF16 \u00e9 utilizado na fase de forma\u00e7\u00e3o, FP8 \u00e9 utilizado na fase de gera\u00e7\u00e3o (o que permite um aumento de velocidade de 1,8x) e o estado do optimizador \u00e9 mantido em FP32<\/li>\n\n\n\n<li><strong>Gera\u00e7\u00e3o de precis\u00e3o FP8<\/strong>O investigador desenvolveu um modelo de gera\u00e7\u00e3o de precis\u00e3o FP8 em linha que suporta a estrutura vLLM, com um d\u00e9bito de gera\u00e7\u00e3o de at\u00e9 32 tokens\/s por prompt numa \u00fanica GPU.<\/li>\n\n\n\n<li><strong>Carregador de pesos vLLM personalizado<\/strong>BF16: convers\u00e3o de pesos BF16 para o formato FP8 em tempo de execu\u00e7\u00e3o<\/li>\n<\/ul>\n\n\n\n<p>Com estas optimiza\u00e7\u00f5es, o LN-Ultra atinge um desempenho 4x superior em termos de rendimento de infer\u00eancia do que o DeepSeek-R1, mantendo uma precis\u00e3o superior.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" src=\"https:\/\/school.myshirtai.com\/wp-content\/uploads\/2025\/05\/image-23.png\" alt=\"\" class=\"wp-image-1022\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h-\u6027\u80fd\u5bf9\u6bd4-\u6253\u7834\u53c2\u6570\u91cf\u4e0e\u6027\u80fd\u7684\u7ebf\u6027\u5173\u7cfb\u795e\u8bdd\">Compara\u00e7\u00e3o de desempenho: desfazendo o mito de uma rela\u00e7\u00e3o linear entre o n\u00famero de par\u00e2metros e o desempenho<\/h2>\n\n\n\n<p>Atrav\u00e9s de testes comparativos, a fam\u00edlia de modelos Llama-Nemotron demonstra um desempenho superior para al\u00e9m da sua escala param\u00e9trica:<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>modela\u00e7\u00e3o<\/th><th>GPQA-Diamante<\/th><th>IFEval<\/th><th>LiveCodeBench<\/th><th>Arena Hard<\/th><\/tr><\/thead><tbody><tr><td>LN-Ultra (253B)<\/td><td>76.01<\/td><td>89.45<\/td><td>66.31<\/td><td>85.2<\/td><\/tr><tr><td>Ver Profundidade-R1<\/td><td>71.5<\/td><td>83.3<\/td><td>&#8211;<\/td><td>81.7<\/td><\/tr><tr><td>Lhama 3.1-405B<\/td><td>70.7<\/td><td>88.5<\/td><td>63.3<\/td><td>82.4<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p>Mesmo o LN-Super mais pequeno (49B) teve um bom desempenho, alcan\u00e7ando uma pontua\u00e7\u00e3o elevada de 88,3 no teste Arena Hard, superando modelos propriet\u00e1rios como o Claude 3.5 Sonnet e o GPT-4o-2024-05-13, e superando modelos de c\u00f3digo aberto muito maiores.<\/p>\n\n\n\n<p>Mais notavelmente, na tarefa JudgeBench fora da distribui\u00e7\u00e3o (distinguindo entre respostas de alta qualidade e de baixa qualidade), o LN-Ultra torna-se o modelo de c\u00f3digo aberto com melhor desempenho, superando significativamente o DeepSeek-R1, e perdendo apenas para o modelo propriet\u00e1rio o3-mini(high). Esta \u00e9 uma boa prova da boa capacidade de generaliza\u00e7\u00e3o do modelo.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h-\u5f00\u6e90\u65b0\u683c\u5c40-\u6548\u7387\u4f18\u5148\u65f6\u4ee3\u7684\u5230\u6765\">O novo cen\u00e1rio de c\u00f3digo aberto: o in\u00edcio da era da efici\u00eancia em primeiro lugar<\/h2>\n\n\n\n<p>O lan\u00e7amento da s\u00e9rie Llama-Nemotron marca uma nova fase de desenvolvimento da IA que d\u00e1 prioridade \u00e0 efici\u00eancia e tem um impacto no sector de muitas formas:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>Quebrar as barreiras dos par\u00e2metros<\/strong>Para superar os modelos maiores numa escala mais pequena e desafiar a sabedoria convencional de que \"maior \u00e9 melhor\".<\/li>\n\n\n\n<li><strong>Reduzir o limiar de implanta\u00e7\u00e3o<\/strong>Design arquitet\u00f3nico eficiente para tornar as implementa\u00e7\u00f5es de modelos de grandes dimens\u00f5es acess\u00edveis a mais empresas<\/li>\n\n\n\n<li><strong>Acelerar a inova\u00e7\u00e3o tecnol\u00f3gica<\/strong>Uma estrat\u00e9gia de fonte totalmente aberta acelerar\u00e1 a difus\u00e3o da tecnologia e da inova\u00e7\u00e3o da IA<\/li>\n\n\n\n<li><strong>Promover a investiga\u00e7\u00e3o sobre a efici\u00eancia<\/strong>: motivar mais investigadores a explorar os limites de efici\u00eancia dos grandes modelos<\/li>\n<\/ol>\n\n\n\n<p>\u00c0 medida que a corrida \u00e0 IA entra numa era em que a efici\u00eancia \u00e9 rei, uma s\u00e9rie de inova\u00e7\u00f5es tornadas p\u00fablicas pela s\u00e9rie Llama-Nemotron da NVIDIA - desde os interruptores de infer\u00eancia din\u00e2mica \u00e0 otimiza\u00e7\u00e3o com reconhecimento de hardware, e desde o treino de dados sint\u00e9ticos \u00e0 aprendizagem por refor\u00e7o em grande escala - est\u00e3o preparadas para influenciar a dire\u00e7\u00e3o futura dos grandes modelos.<\/p>\n\n\n\n<p>A import\u00e2ncia desta revela\u00e7\u00e3o tecnol\u00f3gica reside n\u00e3o s\u00f3 no nascimento de uma nova gera\u00e7\u00e3o de modelos de elevada efici\u00eancia, mas tamb\u00e9m no estabelecimento de uma nova refer\u00eancia t\u00e9cnica para toda a ind\u00fastria de IA, que promove a evolu\u00e7\u00e3o cont\u00ednua da tecnologia de IA no sentido de uma maior praticidade e universalidade. Com o apoio de hardware de nova gera\u00e7\u00e3o, como a futura GPU B100, \u00e9 prov\u00e1vel que esta s\u00e9rie de modelos seja apenas o in\u00edcio da revolu\u00e7\u00e3o da efici\u00eancia.<\/p>\n\n\n\n<p><strong>Se quiser utilizar a conta exclusiva paga oficial GPT Plus, Claude Pro, Grok Super, pode contactar a nossa equipa de profissionais (wx: abch891) se n\u00e3o souber como carregar a sua conta.<\/strong><\/p>\n\n\n\n<table style=\"width: 100%;border-collapse: collapse;border: 1px solid #ddd\">\r\n<thead>\r\n<tr style=\"height: 48px;background-color: #f5f5f5\">\r\n<th style=\"width: 50%;height: 48px;border: 1px solid #ddd;padding: 8px\">\r\n<h4 style=\"margin: 0\">Para mais produtos, consultar<\/h4>\r\n<\/th>\r\n<th style=\"width: 50%;height: 48px;border: 1px solid #ddd;padding: 8px\">\r\n<h4 style=\"margin: 0\">Ver mais em<\/h4>\r\n<\/th>\r\n<\/tr>\r\n<\/thead>\r\n<tbody>\r\n<tr style=\"height: 63px\">\r\n<td style=\"width: 50%;height: 63px;border: 1px solid #ddd;padding: 8px\"><a href=\"https:\/\/www.myshirtai.com\/pt\/\" data-linktype=\"2\">ShirtAI - Intelig\u00eancia penetrante<\/a><\/td>\r\n<td style=\"width: 50%;height: 63px;border: 1px solid #ddd;padding: 8px\"><a href=\"https:\/\/www.myshirtai.com\/pt\/archives\/4425\/\" data-linktype=\"2\">O Grande Modelo do AIGC: inaugurando uma era de dupla revolu\u00e7\u00e3o na engenharia e na ci\u00eancia - Penetrating Intelligence<\/a><\/td>\r\n<\/tr>\r\n<tr style=\"height: 61px\">\r\n<td style=\"width: 50%;height: 61px;border: 1px solid #ddd;padding: 8px\"><a href=\"https:\/\/www.myshirtai.com\/pt\/\" data-linktype=\"2\">1:1 Restaura\u00e7\u00e3o de Claude e GPT Site oficial - AI Cloud Native<\/a><\/td>\r\n<td style=\"width: 50%;height: 61px;border: 1px solid #ddd;padding: 8px\"><a href=\"https:\/\/www.bluelsqkj.com\/archives\/2876\" data-linktype=\"2\">Aplica\u00e7\u00e3o de jogos em direto Leitor de visualiza\u00e7\u00e3o de desporto HD global (recomendado) - Blueshirt Technology<\/a><\/td>\r\n<\/tr>\r\n<tr style=\"height: 54px\">\r\n<td style=\"width: 50%;height: 54px;border: 1px solid #ddd;padding: 8px\"><a href=\"https:\/\/api.mygptmeta.com\/\" data-linktype=\"2\">Servi\u00e7o de tr\u00e2nsito baseado na API oficial - API GPTMeta<\/a><\/td>\r\n<td style=\"width: 50%;height: 54px;border: 1px solid #ddd;padding: 8px\"><a href=\"https:\/\/www.zhihu.com\/question\/621055223\/answer\/3633615705\" data-linktype=\"2\">Ajuda, algu\u00e9m pode dar algumas dicas sobre como fazer perguntas no GPT? - Conhecimento<\/a><\/td>\r\n<\/tr>\r\n<tr style=\"height: 70px\">\r\n<td style=\"width: 50%;height: 70px;border: 1px solid #ddd;padding: 8px\"><a href=\"https:\/\/shop.blueshirtmap.com\/\" data-linktype=\"2\">Loja digital de bens virtuais globais - Global SmarTone (Feng Ling Ge)<\/a><\/td>\r\n<td style=\"width: 50%;height: 70px;border: 1px solid #ddd;padding: 8px\"><a href=\"https:\/\/www.bilibili.com\/video\/BV1efpneYE54\/?spm_id_from=333.1387.homepage.video_card.click\" data-linktype=\"2\">Qu\u00e3o poderosa \u00e9 a funcionalidade Claude airtfacts que o GPT instantaneamente n\u00e3o cheira bem? -BeepBeep<\/a><\/td>\r\n<\/tr>\r\n<\/tbody>\r\n<\/table>","protected":false},"excerpt":{"rendered":"<p>A NVIDIA lan\u00e7a modelos Llama-NemotronAI de c\u00f3digo aberto nas vers\u00f5es 8B, 49B e 253B. O emblem\u00e1tico LN-Ultra supera o DeepSeek-R1 de 671 mil milh\u00f5es em v\u00e1rios par\u00e2metros de refer\u00eancia com apenas 253 mil milh\u00f5es de par\u00e2metros, ao mesmo tempo que permite um funcionamento mais eficiente num \u00fanico n\u00f3 xH100. O processo de forma\u00e7\u00e3o em cinco fases da s\u00e9rie, com t\u00e9cnicas inovadoras, inclui comuta\u00e7\u00e3o de infer\u00eancia, otimiza\u00e7\u00e3o com reconhecimento de hardware e forma\u00e7\u00e3o de dados sint\u00e9ticos. A rela\u00e7\u00e3o positiva entre a escala e o desempenho dos par\u00e2metros de desempenho do modelo marca a era da efici\u00eancia da IA em primeiro lugar, e o seu licenciamento de c\u00f3digo aberto ir\u00e1 acelerar a ado\u00e7\u00e3o da tecnologia.<\/p>","protected":false},"author":1,"featured_media":5690,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_uag_custom_page_level_css":"","site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[53],"tags":[67,68],"class_list":["post-6388","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blog","tag-ai-models","tag-deep-learning"],"yoast_head":"<!-- This site is optimized with the Yoast SEO Premium plugin v22.3 (Yoast SEO v25.2) - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>NVIDIA Llama-Nemotron\uff1a\u8d85\u8d8aDeepSeek-R1\u7684\u5f00\u6e90\u65b0\u738b\u8005 - \u6e17\u900f\u667a\u80fd<\/title>\n<meta name=\"description\" content=\"\u82f1\u4f1f\u8fbe\u53d1\u5e03\u5f00\u6e90Llama-NemotronAI\u6a21\u578b\uff0c\u5305\u62ec8B\u300149B\u548c253B\u7248\u672c\u3002\u65d7\u8230\u7ea7LN-Ultra\u4ee5\u4ec5253\u4ebf\u53c2\u6570\u5728\u591a\u9879\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8d85\u8d8a\u4e866710\u4ebfDeepSeek-R1\uff0c\u540c\u65f6\u5728\u5355\u4e2axH100\u8282\u70b9\u4e0a\u5b9e\u73b0\u66f4\u9ad8\u6548\u8fd0\u884c\u3002\u8be5\u7cfb\u5217\u4e94\u9636\u6bb5\u8bad\u7ec3\u6d41\u7a0b\uff0c\u521b\u65b0\u6027\u6280\u672f\u5305\u62ec\u63a8\u7406\u5f00\u5173\u3001\u786c\u4ef6\u611f\u77e5\u4f18\u5316\u548c\u5408\u6210\u6570\u636e\u8bad\u7ec3\u3002\u6a21\u578b\u6027\u80fd\u53c2\u6570\u89c4\u6a21\u4e0e\u6027\u80fd\u7684\u6b63\u6bd4\u5173\u7cfb\uff0c\u6807\u5fd7\u7740AI\u6548\u7387\u4f18\u5148\u65f6\u4ee3\uff0c\u5176\u5f00\u6e90\u8bb8\u53ef\u5c06\u52a0\u901f\u6280\u672f\u666e\u53ca\u3002\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.myshirtai.com\/pt\/archives\/6388\/\" \/>\n<meta property=\"og:locale\" content=\"pt_PT\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"NVIDIA Llama-Nemotron\uff1a\u8d85\u8d8aDeepSeek-R1\u7684\u5f00\u6e90\u65b0\u738b\u8005\" \/>\n<meta property=\"og:description\" content=\"\u82f1\u4f1f\u8fbe\u53d1\u5e03\u5f00\u6e90Llama-NemotronAI\u6a21\u578b\uff0c\u5305\u62ec8B\u300149B\u548c253B\u7248\u672c\u3002\u65d7\u8230\u7ea7LN-Ultra\u4ee5\u4ec5253\u4ebf\u53c2\u6570\u5728\u591a\u9879\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8d85\u8d8a\u4e866710\u4ebfDeepSeek-R1\uff0c\u540c\u65f6\u5728\u5355\u4e2axH100\u8282\u70b9\u4e0a\u5b9e\u73b0\u66f4\u9ad8\u6548\u8fd0\u884c\u3002\u8be5\u7cfb\u5217\u4e94\u9636\u6bb5\u8bad\u7ec3\u6d41\u7a0b\uff0c\u521b\u65b0\u6027\u6280\u672f\u5305\u62ec\u63a8\u7406\u5f00\u5173\u3001\u786c\u4ef6\u611f\u77e5\u4f18\u5316\u548c\u5408\u6210\u6570\u636e\u8bad\u7ec3\u3002\u6a21\u578b\u6027\u80fd\u53c2\u6570\u89c4\u6a21\u4e0e\u6027\u80fd\u7684\u6b63\u6bd4\u5173\u7cfb\uff0c\u6807\u5fd7\u7740AI\u6548\u7387\u4f18\u5148\u65f6\u4ee3\uff0c\u5176\u5f00\u6e90\u8bb8\u53ef\u5c06\u52a0\u901f\u6280\u672f\u666e\u53ca\u3002\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.myshirtai.com\/pt\/archives\/6388\/\" \/>\n<meta property=\"og:site_name\" content=\"\u6e17\u900f\u667a\u80fd\" \/>\n<meta property=\"article:published_time\" content=\"2025-05-08T03:35:21+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png\" \/>\n\t<meta property=\"og:image:width\" content=\"501\" \/>\n\t<meta property=\"og:image:height\" content=\"900\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"IvesFeng666\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Escrito por\" \/>\n\t<meta name=\"twitter:data1\" content=\"IvesFeng666\" \/>\n\t<meta name=\"twitter:label2\" content=\"Tempo estimado de leitura\" \/>\n\t<meta name=\"twitter:data2\" content=\"3 minutos\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/www.myshirtai.com\/archives\/6388#article\",\"isPartOf\":{\"@id\":\"https:\/\/www.myshirtai.com\/archives\/6388\"},\"author\":{\"name\":\"IvesFeng666\",\"@id\":\"https:\/\/www.myshirtai.com\/#\/schema\/person\/793ffae65b0212a937f22250e83b51e2\"},\"headline\":\"NVIDIA Llama-Nemotron\uff1a\u8d85\u8d8aDeepSeek-R1\u7684\u5f00\u6e90\u65b0\u738b\u8005\",\"datePublished\":\"2025-05-08T03:35:21+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/www.myshirtai.com\/archives\/6388\"},\"wordCount\":169,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\/\/www.myshirtai.com\/#organization\"},\"image\":{\"@id\":\"https:\/\/www.myshirtai.com\/archives\/6388#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png\",\"keywords\":[\"AI\u6a21\u578b\",\"\u6df1\u5ea6\u5b66\u4e60\"],\"articleSection\":[\"\u535a\u5ba2\"],\"inLanguage\":\"pt-PT\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\/\/www.myshirtai.com\/archives\/6388#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.myshirtai.com\/archives\/6388\",\"url\":\"https:\/\/www.myshirtai.com\/archives\/6388\",\"name\":\"NVIDIA Llama-Nemotron\uff1a\u8d85\u8d8aDeepSeek-R1\u7684\u5f00\u6e90\u65b0\u738b\u8005 - \u6e17\u900f\u667a\u80fd\",\"isPartOf\":{\"@id\":\"https:\/\/www.myshirtai.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/www.myshirtai.com\/archives\/6388#primaryimage\"},\"image\":{\"@id\":\"https:\/\/www.myshirtai.com\/archives\/6388#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png\",\"datePublished\":\"2025-05-08T03:35:21+00:00\",\"description\":\"\u82f1\u4f1f\u8fbe\u53d1\u5e03\u5f00\u6e90Llama-NemotronAI\u6a21\u578b\uff0c\u5305\u62ec8B\u300149B\u548c253B\u7248\u672c\u3002\u65d7\u8230\u7ea7LN-Ultra\u4ee5\u4ec5253\u4ebf\u53c2\u6570\u5728\u591a\u9879\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8d85\u8d8a\u4e866710\u4ebfDeepSeek-R1\uff0c\u540c\u65f6\u5728\u5355\u4e2axH100\u8282\u70b9\u4e0a\u5b9e\u73b0\u66f4\u9ad8\u6548\u8fd0\u884c\u3002\u8be5\u7cfb\u5217\u4e94\u9636\u6bb5\u8bad\u7ec3\u6d41\u7a0b\uff0c\u521b\u65b0\u6027\u6280\u672f\u5305\u62ec\u63a8\u7406\u5f00\u5173\u3001\u786c\u4ef6\u611f\u77e5\u4f18\u5316\u548c\u5408\u6210\u6570\u636e\u8bad\u7ec3\u3002\u6a21\u578b\u6027\u80fd\u53c2\u6570\u89c4\u6a21\u4e0e\u6027\u80fd\u7684\u6b63\u6bd4\u5173\u7cfb\uff0c\u6807\u5fd7\u7740AI\u6548\u7387\u4f18\u5148\u65f6\u4ee3\uff0c\u5176\u5f00\u6e90\u8bb8\u53ef\u5c06\u52a0\u901f\u6280\u672f\u666e\u53ca\u3002\",\"breadcrumb\":{\"@id\":\"https:\/\/www.myshirtai.com\/archives\/6388#breadcrumb\"},\"inLanguage\":\"pt-PT\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.myshirtai.com\/archives\/6388\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"pt-PT\",\"@id\":\"https:\/\/www.myshirtai.com\/archives\/6388#primaryimage\",\"url\":\"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png\",\"contentUrl\":\"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png\",\"width\":501,\"height\":900},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.myshirtai.com\/archives\/6388#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.myshirtai.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"NVIDIA Llama-Nemotron\uff1a\u8d85\u8d8aDeepSeek-R1\u7684\u5f00\u6e90\u65b0\u738b\u8005\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.myshirtai.com\/#website\",\"url\":\"https:\/\/www.myshirtai.com\/\",\"name\":\"\u6e17\u900f\u667a\u80fd\",\"description\":\"ShirtAI\",\"publisher\":{\"@id\":\"https:\/\/www.myshirtai.com\/#organization\"},\"alternateName\":\"ShirtAI\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.myshirtai.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"pt-PT\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/www.myshirtai.com\/#organization\",\"name\":\"ShirtAI\",\"alternateName\":\"ShirtAI\",\"url\":\"https:\/\/www.myshirtai.com\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"pt-PT\",\"@id\":\"https:\/\/www.myshirtai.com\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2023\/11\/ShirtAI1279\u00d7675.png\",\"contentUrl\":\"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2023\/11\/ShirtAI1279\u00d7675.png\",\"width\":1200,\"height\":675,\"caption\":\"ShirtAI\"},\"image\":{\"@id\":\"https:\/\/www.myshirtai.com\/#\/schema\/logo\/image\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.myshirtai.com\/#\/schema\/person\/793ffae65b0212a937f22250e83b51e2\",\"name\":\"IvesFeng666\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"pt-PT\",\"@id\":\"https:\/\/www.myshirtai.com\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/0e40122f3ea588c331477d2b5778ab521f0ef9275880700b47f592c999e721b7?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/0e40122f3ea588c331477d2b5778ab521f0ef9275880700b47f592c999e721b7?s=96&d=mm&r=g\",\"caption\":\"IvesFeng666\"},\"sameAs\":[\"http:\/\/www.myshirtai.com\"],\"url\":\"https:\/\/www.myshirtai.com\/pt\/archives\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"NVIDIA Llama-Nemotron\uff1a\u8d85\u8d8aDeepSeek-R1\u7684\u5f00\u6e90\u65b0\u738b\u8005 - \u6e17\u900f\u667a\u80fd","description":"\u82f1\u4f1f\u8fbe\u53d1\u5e03\u5f00\u6e90Llama-NemotronAI\u6a21\u578b\uff0c\u5305\u62ec8B\u300149B\u548c253B\u7248\u672c\u3002\u65d7\u8230\u7ea7LN-Ultra\u4ee5\u4ec5253\u4ebf\u53c2\u6570\u5728\u591a\u9879\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8d85\u8d8a\u4e866710\u4ebfDeepSeek-R1\uff0c\u540c\u65f6\u5728\u5355\u4e2axH100\u8282\u70b9\u4e0a\u5b9e\u73b0\u66f4\u9ad8\u6548\u8fd0\u884c\u3002\u8be5\u7cfb\u5217\u4e94\u9636\u6bb5\u8bad\u7ec3\u6d41\u7a0b\uff0c\u521b\u65b0\u6027\u6280\u672f\u5305\u62ec\u63a8\u7406\u5f00\u5173\u3001\u786c\u4ef6\u611f\u77e5\u4f18\u5316\u548c\u5408\u6210\u6570\u636e\u8bad\u7ec3\u3002\u6a21\u578b\u6027\u80fd\u53c2\u6570\u89c4\u6a21\u4e0e\u6027\u80fd\u7684\u6b63\u6bd4\u5173\u7cfb\uff0c\u6807\u5fd7\u7740AI\u6548\u7387\u4f18\u5148\u65f6\u4ee3\uff0c\u5176\u5f00\u6e90\u8bb8\u53ef\u5c06\u52a0\u901f\u6280\u672f\u666e\u53ca\u3002","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.myshirtai.com\/pt\/archives\/6388\/","og_locale":"pt_PT","og_type":"article","og_title":"NVIDIA Llama-Nemotron\uff1a\u8d85\u8d8aDeepSeek-R1\u7684\u5f00\u6e90\u65b0\u738b\u8005","og_description":"\u82f1\u4f1f\u8fbe\u53d1\u5e03\u5f00\u6e90Llama-NemotronAI\u6a21\u578b\uff0c\u5305\u62ec8B\u300149B\u548c253B\u7248\u672c\u3002\u65d7\u8230\u7ea7LN-Ultra\u4ee5\u4ec5253\u4ebf\u53c2\u6570\u5728\u591a\u9879\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8d85\u8d8a\u4e866710\u4ebfDeepSeek-R1\uff0c\u540c\u65f6\u5728\u5355\u4e2axH100\u8282\u70b9\u4e0a\u5b9e\u73b0\u66f4\u9ad8\u6548\u8fd0\u884c\u3002\u8be5\u7cfb\u5217\u4e94\u9636\u6bb5\u8bad\u7ec3\u6d41\u7a0b\uff0c\u521b\u65b0\u6027\u6280\u672f\u5305\u62ec\u63a8\u7406\u5f00\u5173\u3001\u786c\u4ef6\u611f\u77e5\u4f18\u5316\u548c\u5408\u6210\u6570\u636e\u8bad\u7ec3\u3002\u6a21\u578b\u6027\u80fd\u53c2\u6570\u89c4\u6a21\u4e0e\u6027\u80fd\u7684\u6b63\u6bd4\u5173\u7cfb\uff0c\u6807\u5fd7\u7740AI\u6548\u7387\u4f18\u5148\u65f6\u4ee3\uff0c\u5176\u5f00\u6e90\u8bb8\u53ef\u5c06\u52a0\u901f\u6280\u672f\u666e\u53ca\u3002","og_url":"https:\/\/www.myshirtai.com\/pt\/archives\/6388\/","og_site_name":"\u6e17\u900f\u667a\u80fd","article_published_time":"2025-05-08T03:35:21+00:00","og_image":[{"width":501,"height":900,"url":"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png","type":"image\/png"}],"author":"IvesFeng666","twitter_card":"summary_large_image","twitter_misc":{"Escrito por":"IvesFeng666","Tempo estimado de leitura":"3 minutos"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.myshirtai.com\/archives\/6388#article","isPartOf":{"@id":"https:\/\/www.myshirtai.com\/archives\/6388"},"author":{"name":"IvesFeng666","@id":"https:\/\/www.myshirtai.com\/#\/schema\/person\/793ffae65b0212a937f22250e83b51e2"},"headline":"NVIDIA Llama-Nemotron\uff1a\u8d85\u8d8aDeepSeek-R1\u7684\u5f00\u6e90\u65b0\u738b\u8005","datePublished":"2025-05-08T03:35:21+00:00","mainEntityOfPage":{"@id":"https:\/\/www.myshirtai.com\/archives\/6388"},"wordCount":169,"commentCount":0,"publisher":{"@id":"https:\/\/www.myshirtai.com\/#organization"},"image":{"@id":"https:\/\/www.myshirtai.com\/archives\/6388#primaryimage"},"thumbnailUrl":"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png","keywords":["AI\u6a21\u578b","\u6df1\u5ea6\u5b66\u4e60"],"articleSection":["\u535a\u5ba2"],"inLanguage":"pt-PT","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/www.myshirtai.com\/archives\/6388#respond"]}]},{"@type":"WebPage","@id":"https:\/\/www.myshirtai.com\/archives\/6388","url":"https:\/\/www.myshirtai.com\/archives\/6388","name":"NVIDIA Llama-Nemotron\uff1a\u8d85\u8d8aDeepSeek-R1\u7684\u5f00\u6e90\u65b0\u738b\u8005 - \u6e17\u900f\u667a\u80fd","isPartOf":{"@id":"https:\/\/www.myshirtai.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.myshirtai.com\/archives\/6388#primaryimage"},"image":{"@id":"https:\/\/www.myshirtai.com\/archives\/6388#primaryimage"},"thumbnailUrl":"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png","datePublished":"2025-05-08T03:35:21+00:00","description":"\u82f1\u4f1f\u8fbe\u53d1\u5e03\u5f00\u6e90Llama-NemotronAI\u6a21\u578b\uff0c\u5305\u62ec8B\u300149B\u548c253B\u7248\u672c\u3002\u65d7\u8230\u7ea7LN-Ultra\u4ee5\u4ec5253\u4ebf\u53c2\u6570\u5728\u591a\u9879\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8d85\u8d8a\u4e866710\u4ebfDeepSeek-R1\uff0c\u540c\u65f6\u5728\u5355\u4e2axH100\u8282\u70b9\u4e0a\u5b9e\u73b0\u66f4\u9ad8\u6548\u8fd0\u884c\u3002\u8be5\u7cfb\u5217\u4e94\u9636\u6bb5\u8bad\u7ec3\u6d41\u7a0b\uff0c\u521b\u65b0\u6027\u6280\u672f\u5305\u62ec\u63a8\u7406\u5f00\u5173\u3001\u786c\u4ef6\u611f\u77e5\u4f18\u5316\u548c\u5408\u6210\u6570\u636e\u8bad\u7ec3\u3002\u6a21\u578b\u6027\u80fd\u53c2\u6570\u89c4\u6a21\u4e0e\u6027\u80fd\u7684\u6b63\u6bd4\u5173\u7cfb\uff0c\u6807\u5fd7\u7740AI\u6548\u7387\u4f18\u5148\u65f6\u4ee3\uff0c\u5176\u5f00\u6e90\u8bb8\u53ef\u5c06\u52a0\u901f\u6280\u672f\u666e\u53ca\u3002","breadcrumb":{"@id":"https:\/\/www.myshirtai.com\/archives\/6388#breadcrumb"},"inLanguage":"pt-PT","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.myshirtai.com\/archives\/6388"]}]},{"@type":"ImageObject","inLanguage":"pt-PT","@id":"https:\/\/www.myshirtai.com\/archives\/6388#primaryimage","url":"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png","contentUrl":"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png","width":501,"height":900},{"@type":"BreadcrumbList","@id":"https:\/\/www.myshirtai.com\/archives\/6388#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.myshirtai.com\/"},{"@type":"ListItem","position":2,"name":"NVIDIA Llama-Nemotron\uff1a\u8d85\u8d8aDeepSeek-R1\u7684\u5f00\u6e90\u65b0\u738b\u8005"}]},{"@type":"WebSite","@id":"https:\/\/www.myshirtai.com\/#website","url":"https:\/\/www.myshirtai.com\/","name":"\u6e17\u900f\u667a\u80fd","description":"ShirtAI","publisher":{"@id":"https:\/\/www.myshirtai.com\/#organization"},"alternateName":"ShirtAI","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.myshirtai.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"pt-PT"},{"@type":"Organization","@id":"https:\/\/www.myshirtai.com\/#organization","name":"ShirtAI","alternateName":"ShirtAI","url":"https:\/\/www.myshirtai.com\/","logo":{"@type":"ImageObject","inLanguage":"pt-PT","@id":"https:\/\/www.myshirtai.com\/#\/schema\/logo\/image\/","url":"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2023\/11\/ShirtAI1279\u00d7675.png","contentUrl":"https:\/\/www.myshirtai.com\/wp-content\/uploads\/2023\/11\/ShirtAI1279\u00d7675.png","width":1200,"height":675,"caption":"ShirtAI"},"image":{"@id":"https:\/\/www.myshirtai.com\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/www.myshirtai.com\/#\/schema\/person\/793ffae65b0212a937f22250e83b51e2","name":"IvesFeng666","image":{"@type":"ImageObject","inLanguage":"pt-PT","@id":"https:\/\/www.myshirtai.com\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/0e40122f3ea588c331477d2b5778ab521f0ef9275880700b47f592c999e721b7?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/0e40122f3ea588c331477d2b5778ab521f0ef9275880700b47f592c999e721b7?s=96&d=mm&r=g","caption":"IvesFeng666"},"sameAs":["http:\/\/www.myshirtai.com"],"url":"https:\/\/www.myshirtai.com\/pt\/archives\/author\/admin"}]}},"uagb_featured_image_src":{"full":["https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png",501,900,false],"thumbnail":["https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20-150x150.png",150,150,true],"medium":["https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20-167x300.png",167,300,true],"medium_large":["https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png",501,900,false],"large":["https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png",501,900,false],"1536x1536":["https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png",501,900,false],"2048x2048":["https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20.png",501,900,false],"trp-custom-language-flag":["https:\/\/www.myshirtai.com\/wp-content\/uploads\/2025\/02\/image-20-7x12.png",7,12,true]},"uagb_author_info":{"display_name":"IvesFeng666","author_link":"https:\/\/www.myshirtai.com\/pt\/archives\/author\/admin"},"uagb_comment_info":0,"uagb_excerpt":"\u82f1\u4f1f\u8fbe\u53d1\u5e03\u5f00\u6e90Llama-NemotronAI\u6a21\u578b\uff0c\u5305\u62ec8B\u300149B\u548c253B\u7248\u672c\u3002\u65d7\u8230\u7ea7LN-Ultra\u4ee5\u4ec5&hellip;","_links":{"self":[{"href":"https:\/\/www.myshirtai.com\/pt\/wp-json\/wp\/v2\/posts\/6388","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.myshirtai.com\/pt\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.myshirtai.com\/pt\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.myshirtai.com\/pt\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.myshirtai.com\/pt\/wp-json\/wp\/v2\/comments?post=6388"}],"version-history":[{"count":0,"href":"https:\/\/www.myshirtai.com\/pt\/wp-json\/wp\/v2\/posts\/6388\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.myshirtai.com\/pt\/wp-json\/wp\/v2\/media\/5690"}],"wp:attachment":[{"href":"https:\/\/www.myshirtai.com\/pt\/wp-json\/wp\/v2\/media?parent=6388"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.myshirtai.com\/pt\/wp-json\/wp\/v2\/categories?post=6388"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.myshirtai.com\/pt\/wp-json\/wp\/v2\/tags?post=6388"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}