{"id":5277,"date":"2025-04-07T03:54:43","date_gmt":"2025-04-06T19:54:43","guid":{"rendered":"https:\/\/infernews.com\/?p=5277"},"modified":"2025-04-07T05:58:31","modified_gmt":"2025-04-06T21:58:31","slug":"llama-4-%e6%a8%a1%e5%9e%8b%e7%b3%bb%e5%88%97","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/llama-4-%e6%a8%a1%e5%9e%8b%e7%b3%bb%e5%88%97\/","title":{"rendered":"Llama 4 \u6a21\u578b\u7cfb\u5217"},"content":{"rendered":"\n<p><a href=\"https:\/\/ai.meta.com\/blog\/llama-4-multimodal-intelligence\/\" target=\"_blank\" rel=\"noreferrer noopener\">Llama 4<\/a> Scout \u662f\u4e00\u500b\u64c1\u6709 170 \u5104\u500b\u6d3b\u8e8d\u53c3\u6578\u548c 16 \u500b MOE \u7684\u6df7\u5408\u5c08\u5bb6\u6a21\u578b\u3002\u5b83\u88ab\u8a8d\u70ba\u662f\u76ee\u524d\u540c\u985e\u578b\u6700\u512a\u79c0\u7684\u591a\u6a21\u614b\u6a21\u578b\uff0c\u6bd4\u524d\u5e7e\u4ee3\u7684 Llama \u6a21\u578b\u66f4\u5f37\u5927\u3002\u65b0\u6a21\u578b\u53ef\u4ee5\u7528\u55ae\u5f35 NVIDIA H100 GPU \u904b\u4f5c\u3002\u4f62\u64c1\u6709\u696d\u4ecb\u9818\u5148\u7684 1000 \u842c tokens \u4e0a\u4e0b\u6587\u7a97\u53e3\uff0c\u4e26\u4e14\u5728\u5ee3\u6cdb\u4f7f\u7528\u7684\u57fa\u6e96\u6e2c\u8a66\u8868\u73fe\u90fd\u512a\u52dd\u904e Gemma 3\u3001Gemini 2.0 Flash-Lite \u548c Mistral 3.1\u3002\u5b83\u5728\u9810\u8a13\u7df4\u548c\u5f8c\u8a13\u7df4\u7684\u904e\u7a0b\u90fd\u7528\u4e86 256K \u7684\u4e0a\u4e0b\u6587\u9577\u5ea6\u3002<\/p>\n\n\n\n<p>\u53e6\u4e00\u500b\u6a21\u578b Maverick \u540c\u6a23\u662f\u4e00\u500b\u64c1\u6709 170 \u5104\u500b\u6d3b\u8e8d\u53c3\u6578\u7684\u6a21\u578b\uff0c\u4f46\u5b83\u64c1\u6709 128 \u500b MOE \u7684\u6df7\u5408\u5c08\u5bb6\u6a21\u578b\u3002\u57fa\u6e96\u6e2c\u8a66\u64ca\u6557\u4e86 GPT-4o \u548c Gemini 2.0 Flash\uff0c\u800c\u5728\u5716\u50cf\u65b9\u9762\u4ea6\u8868\u73fe\u76f8\u7576\u51fa\u8272\uff0c\u80fd\u5920\u5c07\u63d0\u793a\u8a5e\u540c\u57cb\u76f8\u95dc\u7684\u8996\u89ba\u6982\u5ff5\u5c0d\u9f4a\uff0c\u5c07\u6a21\u578b\u7684\u56de\u61c9\u9396\u5b9a\u5230\u5716\u50cf\u4e2d\u7684\u7279\u5b9a\u5340\u57df\u3002\u5169\u500b\u6a21\u578b\u90fd\u6709\u7368\u7279\u7684\u6d3b\u8e8d\u53c3\u6578\u6a21\u5f0f\uff0c\u80fd\u5920\u7bc0\u7701\u4e00\u534a\u8cc7\u6e90\u3002\u4ee4\u958b\u767c\u540c\u61c9\u7528\u7684\u50f9\u683c\u66f4\u4f4e\u3002<\/p>\n\n\n\n<figure class=\"wp-block-audio\"><audio controls src=\"\/blog\/wp-content\/uploads\/2025\/04\/1743969044441841681-255656878002442.mp3\"><\/audio><\/figure>\n\n\n\n<figure class=\"wp-block-image aligncenter size-large\"><a href=\"https:\/\/ai.meta.com\/blog\/llama-4-multimodal-intelligence\/\" target=\"_blank\" rel=\" noreferrer noopener\"><img data-dominant-color=\"b7ccdf\" data-has-transparency=\"false\" style=\"--dominant-color: #b7ccdf;\" loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"535\" src=\"\/blog\/wp-content\/uploads\/2025\/04\/489528324_1866126614188079_2353760794201377773_n-1024x535.jpg\" alt=\"\" class=\"wp-image-5278 not-transparent\" srcset=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2025\/04\/489528324_1866126614188079_2353760794201377773_n-1024x535.jpg 1024w, https:\/\/infernews.com\/blog\/wp-content\/uploads\/2025\/04\/489528324_1866126614188079_2353760794201377773_n-300x157.jpg 300w, https:\/\/infernews.com\/blog\/wp-content\/uploads\/2025\/04\/489528324_1866126614188079_2353760794201377773_n-768x401.jpg 768w, https:\/\/infernews.com\/blog\/wp-content\/uploads\/2025\/04\/489528324_1866126614188079_2353760794201377773_n.jpg 1080w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/a><\/figure>\n\n\n<figure class=\"wp-block-embed-youtube wp-block-embed is-type-video is-provider-youtube wp-embed-aspect-16-9 wp-has-aspect-ratio\"><div class=\"lyte-wrapper\" title=\"Llama-4 First Look &amp;amp; Hands On Testing (Code, Vision, Dialogue)\" style=\"width:853px;max-width:100%;margin:5px auto;\"><div class=\"lyMe\" id=\"WYL_HV_B7Ak9GVk\" itemprop=\"video\" itemscope itemtype=\"https:\/\/schema.org\/VideoObject\"><div><meta itemprop=\"thumbnailUrl\" content=\"https:\/\/infernews.com\/blog\/wp-content\/plugins\/wp-youtube-lyte\/lyteCache.php?origThumbUrl=https%3A%2F%2Fi.ytimg.com%2Fvi%2FHV_B7Ak9GVk%2Fhqdefault.jpg\" \/><meta itemprop=\"embedURL\" content=\"https:\/\/www.youtube.com\/embed\/HV_B7Ak9GVk\" \/><meta itemprop=\"duration\" content=\"PT16M3S\" \/><meta itemprop=\"uploadDate\" content=\"2025-04-06T05:10:55Z\" \/><\/div><div id=\"lyte_HV_B7Ak9GVk\" data-src=\"https:\/\/infernews.com\/blog\/wp-content\/plugins\/wp-youtube-lyte\/lyteCache.php?origThumbUrl=https%3A%2F%2Fi.ytimg.com%2Fvi%2FHV_B7Ak9GVk%2Fhqdefault.jpg\" class=\"pL\"><div class=\"tC\"><div class=\"tT\" itemprop=\"name\">Llama-4 First Look &amp; Hands On Testing (Code, Vision, Dialogue)<\/div><\/div><div class=\"play\"><\/div><div class=\"ctrl\"><div class=\"Lctrl\"><\/div><div class=\"Rctrl\"><\/div><\/div><\/div><noscript><a href=\"https:\/\/youtu.be\/HV_B7Ak9GVk\" rel=\"nofollow\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/plugins\/wp-youtube-lyte\/lyteCache.php?origThumbUrl=https%3A%2F%2Fi.ytimg.com%2Fvi%2FHV_B7Ak9GVk%2F0.jpg\" alt=\"Llama-4 First Look &amp;amp; Hands On Testing (Code, Vision, Dialogue)\" width=\"853\" height=\"460\" \/><br \/>Watch this video on YouTube<\/a><\/noscript><meta itemprop=\"description\" content=\"Timestamps: 00:00 - First Look 04:39 - Python Game Test 07:21 - Game Test Two 08:45 - Roleplay Testing 09:30 - Hallucination Test 10:18 - Image Gen Test 11:11 - Image Animation Test 12:53 - OCR Test 14:17 - Closing Thoughts In this video, we take a first hands-on look at Meta\u2019s newly released Llama 4 models\u2014a highly anticipated update to the Llama family. This release introduces three new MoE (Mixture of Experts) models, each larger and more capable, though not yet practical to run locally without quantization or significant hardware. We begin with a quick overview of the announcement details, then move into a series of real-world tests to see how Llama 4 performs in creative and reasoning tasks. The testing includes: Python-based game generation Roleplay and narrative exploration Hallucination probing Image generation Image animation OCR and text recognition tasks Rather than diving deep into benchmarks or model specs, this video focuses on interaction and practical use, aiming to get a feel for the model\u2019s personality, creativity, and reasoning abilities through casual experimentation.\"><\/div><\/div><div class=\"lL\" style=\"max-width:100%;width:853px;margin:5px auto;\"><\/div><figcaption><\/figcaption><\/figure>","protected":false},"excerpt":{"rendered":"<p>Llama 4 Scout \u662f\u4e00\u500b\u64c1\u6709 170 \u5104\u500b\u6d3b\u8e8d\u53c3\u6578\u548c 16 \u500b MOE \u7684\u6df7\u5408\u5c08\u5bb6\u6a21\u578b\u3002\u5b83\u88ab\u8a8d\u70ba\u662f\u76ee\u524d\u540c\u985e\u578b\u6700\u512a\u79c0\u7684\u591a\u6a21\u614b\u6a21\u578b\uff0c\u6bd4\u524d\u5e7e\u4ee3\u7684 Llama \u6a21\u578b\u66f4\u5f37\u5927\u3002\u65b0\u6a21\u578b\u53ef\u4ee5\u7528\u55ae\u5f35 NVIDIA H100 GPU \u904b\u4f5c\u3002\u4f62\u64c1\u6709\u696d\u4ecb\u9818\u5148\u7684 1000 \u842c tokens \u4e0a\u4e0b\u6587\u7a97\u53e3\uff0c\u4e26\u4e14\u5728\u5ee3\u6cdb\u4f7f\u7528\u7684\u57fa\u6e96\u6e2c\u8a66\u8868\u73fe\u90fd\u512a\u52dd\u904e Gemma 3\u3001Gemini 2.0 Flash-Lite \u548c Mistral 3.1\u3002\u5b83\u5728\u9810\u8a13\u7df4\u548c\u5f8c\u8a13\u7df4\u7684\u904e\u7a0b\u90fd\u7528\u4e86 256K \u7684\u4e0a\u4e0b\u6587\u9577\u5ea6\u3002 \u53e6\u4e00\u500b\u6a21\u578b Maverick \u540c\u6a23\u662f\u4e00\u500b\u64c1\u6709 170 \u5104\u500b\u6d3b\u8e8d\u53c3\u6578\u7684\u6a21\u578b\uff0c\u4f46\u5b83\u64c1\u6709 128 \u500b MOE \u7684\u6df7\u5408\u5c08\u5bb6\u6a21\u578b\u3002\u57fa\u6e96\u6e2c\u8a66\u64ca\u6557\u4e86 GPT-4o \u548c Gemini 2.0 Flash\uff0c\u800c\u5728\u5716\u50cf\u65b9\u9762\u4ea6\u8868\u73fe\u76f8\u7576\u51fa\u8272\uff0c\u80fd\u5920\u5c07\u63d0\u793a\u8a5e\u540c\u57cb\u76f8\u95dc\u7684\u8996\u89ba\u6982\u5ff5\u5c0d\u9f4a\uff0c\u5c07\u6a21\u578b\u7684\u56de\u61c9\u9396\u5b9a\u5230\u5716\u50cf\u4e2d\u7684\u7279\u5b9a\u5340\u57df\u3002\u5169\u500b\u6a21\u578b\u90fd\u6709\u7368\u7279\u7684\u6d3b\u8e8d\u53c3\u6578\u6a21\u5f0f\uff0c\u80fd\u5920\u7bc0\u7701\u4e00\u534a\u8cc7\u6e90\u3002\u4ee4\u958b\u767c\u540c\u61c9\u7528\u7684\u50f9\u683c\u66f4\u4f4e\u3002<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"googlesitekit_rrm_CAowvqSiDA:productID":"","footnotes":""},"categories":[40,76,133],"tags":[],"class_list":["post-5277","post","type-post","status-publish","format-standard","hentry","category-llama","category-76","category-133"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/5277","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=5277"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/5277\/revisions"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=5277"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=5277"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=5277"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}