{"id":378995,"date":"2024-06-16T21:00:13","date_gmt":"2024-06-16T21:00:13","guid":{"rendered":"http:\/\/savepearlharbor.com\/?p=378995"},"modified":"-0001-11-30T00:00:00","modified_gmt":"-0001-11-29T21:00:00","slug":"","status":"publish","type":"post","link":"https:\/\/savepearlharbor.com\/?p=378995","title":{"rendered":"<span>Nvidia \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u0438\u043b\u0430 \u043e\u0442\u043a\u0440\u044b\u0442\u0443\u044e LLM \u043c\u043e\u0434\u0435\u043b\u044c Nemotron-4 340B<\/span>"},"content":{"rendered":"<div><!--[--><!--]--><\/div>\n<div id=\"post-content-body\">\n<div>\n<div class=\"article-formatted-body article-formatted-body article-formatted-body_version-2\">\n<div xmlns=\"http:\/\/www.w3.org\/1999\/xhtml\">\n<figure class=\"full-width\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/habrastorage.org\/r\/w1560\/getpro\/habr\/upload_files\/902\/1a5\/e27\/9021a5e278fe3ad10a485d5382c38f53.png\" alt=\"\u0421\u0440\u0430\u0432\u043d\u0435\u043d\u0438\u0435 Nemotron-4 340B \u0441 GPT-4-1106\" title=\"\u0421\u0440\u0430\u0432\u043d\u0435\u043d\u0438\u0435 Nemotron-4 340B \u0441 GPT-4-1106\" width=\"999\" height=\"616\" data-src=\"https:\/\/habrastorage.org\/getpro\/habr\/upload_files\/902\/1a5\/e27\/9021a5e278fe3ad10a485d5382c38f53.png\"\/><\/p>\n<div><figcaption>\u0421\u0440\u0430\u0432\u043d\u0435\u043d\u0438\u0435 Nemotron-4 340B \u0441 GPT-4-1106<\/figcaption><\/div>\n<\/figure>\n<p>\u041d\u0430 \u0434\u0430\u043d\u043d\u044b\u0439 \u043c\u043e\u043c\u0435\u043d\u0442 \u0441\u0430\u043c\u0430\u044f \u043a\u0440\u0443\u043f\u043d\u0430\u044f, \u043f\u043e\u0441\u043b\u0435 DeepSeek-V2 236B, \u043e\u0442\u043a\u0440\u044b\u0442\u0430\u044f LLM \u043c\u043e\u0434\u0435\u043b\u044c \u0440\u0430\u0437\u043c\u0435\u0440\u043e\u043c 340B. \u041f\u043e \u0442\u0435\u0441\u0442\u0430\u043c \u0441\u0440\u0430\u0432\u043d\u0435\u043d\u0438\u044f \u0447\u0435\u043b\u043e\u0432\u0435\u043a\u043e\u043c \u043a\u0430\u043a \u0432\u044b\u0438\u0433\u0440\u044b\u0432\u0430\u0435\u0442 \u0443 GPT-4-1106 (\u0432\u044b\u0448\u043b\u0430 \u0432 \u043d\u043e\u044f\u0431\u0440\u0435 2023), \u0442\u0430\u043a \u0438 \u043f\u0440\u043e\u0438\u0433\u0440\u044b\u0432\u0430\u0435\u0442, \u043d\u043e \u0447\u0430\u0449\u0435 \u0441\u043e\u0441\u0442\u0430\u0432\u043b\u044f\u0435\u0442 \u043f\u0430\u0440\u0438\u0442\u0435\u0442.<\/p>\n<p>\u041e\u0431\u0443\u0447\u0430\u043b\u0430\u0441\u044c \u043d\u0430 50+ \u0435\u0441\u0442\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u0445 \u044f\u0437\u044b\u043a\u0430\u0445 \u0438 40+ \u044f\u0437\u044b\u043a\u0430\u0445 \u043f\u0440\u043e\u0433\u0440\u0430\u043c\u043c\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u044f. \u0410\u0440\u0445\u0438\u0442\u0435\u043a\u0442\u0443\u0440\u0430 \u043c\u043e\u0434\u0435\u043b\u0438 \u043f\u043e\u0441\u0442\u0440\u043e\u0435\u043d\u0430 \u043d\u0430 Grouped-Query Attention (GQA) \u0438 Rotary Position Embeddings (RoPE). \u041d\u043e \u0434\u043b\u0438\u043d\u0430 \u043a\u043e\u043d\u0442\u0435\u043a\u0441\u0442\u0430 \u0432\u0441\u0435\u0433\u043e 4096 \u0442\u043e\u043a\u0435\u043d\u043e\u0432.<\/p>\n<p>\u041c\u043e\u0434\u0435\u043b\u044c \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u043b\u0435\u043d\u0430 \u0432 \u0432\u0438\u0434\u0435: Nemotron-4-340B-Base, Nemotron-4-340B-Instruct \u0438 Nemotron-4-340B-Reward.<\/p>\n<ul>\n<li>\n<p>Base &#8212; \u043f\u0440\u0435\u0434\u043d\u0430\u0437\u043d\u0430\u0447\u0435\u043d\u0430 \u0434\u043b\u044f \u0433\u0435\u043d\u0435\u0440\u0430\u0446\u0438\u0438 \u0441\u0438\u043d\u0442\u0435\u0442\u0438\u0447\u0435\u0441\u043a\u0438\u0445 \u0434\u0430\u043d\u043d\u044b\u0445.<\/p>\n<\/li>\n<li>\n<p>Insctuct &#8212; \u043f\u0440\u0435\u0434\u043d\u0430\u0437\u043d\u0430\u0447\u0435\u043d\u0430 \u0434\u043b\u044f \u0447\u0430\u0442\u0430 \u0438 \u0432\u044b\u043f\u043e\u043b\u043d\u0435\u043d\u0438\u044f \u0438\u043d\u0441\u0442\u0440\u0443\u043a\u0446\u0438\u0439<\/p>\n<\/li>\n<li>\n<p>Reward &#8212; base \u043c\u043e\u0434\u0435\u043b\u044c \u0441 \u0434\u043e\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u044b\u043c \u043b\u0438\u043d\u0435\u0439\u043d\u044b\u043c \u0441\u043b\u043e\u0435\u043c \u0434\u043b\u044f \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u0443\u044f \u043d\u043e\u0432\u044b\u0439 \u043f\u043e\u0434\u0445\u043e\u0434 reward.<\/p>\n<\/li>\n<\/ul>\n<p>\u041c\u043e\u0434\u0435\u043b\u044c \u0440\u0430\u0441\u043f\u0440\u043e\u0441\u0442\u0440\u0430\u043d\u044f\u0435\u0442\u0441\u044f \u043f\u043e\u0434 \u043b\u0438\u0446\u0435\u043d\u0437\u0438\u0435\u0439 NVIDIA Open Model License Agreement, \u0440\u0430\u0437\u0440\u0435\u0448\u0430\u044e\u0449\u0430\u044f \u043a\u043e\u043c\u043c\u0435\u0440\u0447\u0435\u0441\u043a\u043e\u0435 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u043d\u0438\u0435.<\/p>\n<p>\u0421\u0430\u043c\u0430 \u043c\u043e\u0434\u0435\u043b\u044c \u043c\u043e\u0436\u0435\u0442 \u0438 \u0432 \u0441\u0442\u0438\u0445\u0438, \u0442\u0430\u043a \u043a\u0430\u043a \u043e\u0431\u0443\u0447\u0430\u043b\u0430\u0441\u044c \u0438 \u043d\u0430 \u0440\u0443\u0441\u0441\u043a\u043e\u043c \u044f\u0437\u044b\u043a\u0435:<\/p>\n<figure class=\"full-width\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/habrastorage.org\/r\/w1560\/getpro\/habr\/upload_files\/b16\/7f3\/31e\/b167f331e18c32c45475361c2910fb0c.png\" width=\"1357\" height=\"658\" data-src=\"https:\/\/habrastorage.org\/getpro\/habr\/upload_files\/b16\/7f3\/31e\/b167f331e18c32c45475361c2910fb0c.png\"\/><\/figure>\n<p>\u041e\u043d\u043b\u0430\u0439\u043d \u0434\u0435\u043c\u043e: <a href=\"https:\/\/chat.lmsys.org\/\" rel=\"noopener noreferrer nofollow\">https:\/\/chat.lmsys.org\/<\/a> (\u0442\u0430\u043c \u0432\u044b\u0431\u0440\u0430\u0442\u044c Direct Chat)<\/p>\n<p>\u0412\u0435\u0441\u0430 \u043c\u043e\u0434\u0435\u043b\u0438: <a href=\"https:\/\/huggingface.co\/nvidia\/Nemotron-4-340B-Instruct\" rel=\"noopener noreferrer nofollow\">https:\/\/huggingface.co\/nvidia\/Nemotron-4-340B-Instruct<\/a><\/p>\n<p>safetensors: <a href=\"https:\/\/huggingface.co\/failspy\/Nemotron-4-340B-Instruct-SafeTensors\" rel=\"noopener noreferrer nofollow\">https:\/\/huggingface.co\/failspy\/Nemotron-4-340B-Instruct-SafeTensors<\/a><\/p>\n<p>\u0411\u043e\u043b\u044c\u0448\u0435 \u043f\u043e\u0434\u0440\u043e\u0431\u043d\u043e\u0441\u0442\u0435\u0439 \u0432 <a href=\"https:\/\/d1qx31qr3h6wln.cloudfront.net\/publications\/Nemotron_4_340B_8T_0.pdf\" rel=\"noopener noreferrer nofollow\">\u043f\u0440\u0435\u0441\u0441-\u0440\u0435\u043b\u0438\u0437\u0435<\/a>.<\/p>\n<p>\u0421\u0430\u043c\u0430\u044f \u0431\u043e\u043b\u044c\u0448\u0430\u044f \u0437\u0430\u0433\u0430\u0434\u043a\u0430 &#8212; \u0441\u043c\u043e\u0436\u0435\u0442 \u043b\u0438 1 \u0431\u0438\u0442\u043d\u043e\u0435 \u043a\u0432\u0430\u043d\u0442\u043e\u0432\u0430\u043d\u0438\u0435 gguf \u043f\u043e\u0437\u0432\u043e\u043b\u0438\u0442\u044c \u0437\u0430\u043f\u0443\u0441\u0442\u0438\u0442\u044c \u0435\u0451 \u043b\u043e\u043a\u0430\u043b\u044c\u043d\u043e, \u0438 \u0447\u0442\u043e \u0434\u0430\u0441\u0442 \u044d\u0442\u0430 \u043c\u043e\u0434\u0435\u043b\u044c \u0434\u043b\u044f \u0441\u043e\u043e\u0431\u0449\u0435\u0441\u0442\u0432\u0430 llm.<\/p>\n<\/p>\n<\/div>\n<\/div>\n<\/div>\n<p><!----><!----><\/div>\n<p><!----><!----><br \/> \u0441\u0441\u044b\u043b\u043a\u0430 \u043d\u0430 \u043e\u0440\u0438\u0433\u0438\u043d\u0430\u043b \u0441\u0442\u0430\u0442\u044c\u0438 <a href=\"https:\/\/habr.com\/ru\/articles\/822167\/\"> https:\/\/habr.com\/ru\/articles\/822167\/<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<div><!--[--><!--]--><\/div>\n<div id=\"post-content-body\">\n<div>\n<div class=\"article-formatted-body article-formatted-body article-formatted-body_version-2\">\n<div xmlns=\"http:\/\/www.w3.org\/1999\/xhtml\">\n<figure class=\"full-width\">\n<div><figcaption>\u0421\u0440\u0430\u0432\u043d\u0435\u043d\u0438\u0435 Nemotron-4 340B \u0441 GPT-4-1106<\/figcaption><\/div>\n<\/figure>\n<p>\u041d\u0430 \u0434\u0430\u043d\u043d\u044b\u0439 \u043c\u043e\u043c\u0435\u043d\u0442 \u0441\u0430\u043c\u0430\u044f \u043a\u0440\u0443\u043f\u043d\u0430\u044f, \u043f\u043e\u0441\u043b\u0435 DeepSeek-V2 236B, \u043e\u0442\u043a\u0440\u044b\u0442\u0430\u044f LLM \u043c\u043e\u0434\u0435\u043b\u044c \u0440\u0430\u0437\u043c\u0435\u0440\u043e\u043c 340B. \u041f\u043e \u0442\u0435\u0441\u0442\u0430\u043c \u0441\u0440\u0430\u0432\u043d\u0435\u043d\u0438\u044f \u0447\u0435\u043b\u043e\u0432\u0435\u043a\u043e\u043c \u043a\u0430\u043a \u0432\u044b\u0438\u0433\u0440\u044b\u0432\u0430\u0435\u0442 \u0443 GPT-4-1106 (\u0432\u044b\u0448\u043b\u0430 \u0432 \u043d\u043e\u044f\u0431\u0440\u0435 2023), \u0442\u0430\u043a \u0438 \u043f\u0440\u043e\u0438\u0433\u0440\u044b\u0432\u0430\u0435\u0442, \u043d\u043e \u0447\u0430\u0449\u0435 \u0441\u043e\u0441\u0442\u0430\u0432\u043b\u044f\u0435\u0442 \u043f\u0430\u0440\u0438\u0442\u0435\u0442.<\/p>\n<p>\u041e\u0431\u0443\u0447\u0430\u043b\u0430\u0441\u044c \u043d\u0430 50+ \u0435\u0441\u0442\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u0445 \u044f\u0437\u044b\u043a\u0430\u0445 \u0438 40+ \u044f\u0437\u044b\u043a\u0430\u0445 \u043f\u0440\u043e\u0433\u0440\u0430\u043c\u043c\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u044f. \u0410\u0440\u0445\u0438\u0442\u0435\u043a\u0442\u0443\u0440\u0430 \u043c\u043e\u0434\u0435\u043b\u0438 \u043f\u043e\u0441\u0442\u0440\u043e\u0435\u043d\u0430 \u043d\u0430 Grouped-Query Attention (GQA) \u0438 Rotary Position Embeddings (RoPE). \u041d\u043e \u0434\u043b\u0438\u043d\u0430 \u043a\u043e\u043d\u0442\u0435\u043a\u0441\u0442\u0430 \u0432\u0441\u0435\u0433\u043e 4096 \u0442\u043e\u043a\u0435\u043d\u043e\u0432.<\/p>\n<p>\u041c\u043e\u0434\u0435\u043b\u044c \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u043b\u0435\u043d\u0430 \u0432 \u0432\u0438\u0434\u0435: Nemotron-4-340B-Base, Nemotron-4-340B-Instruct \u0438 Nemotron-4-340B-Reward.<\/p>\n<ul>\n<li>\n<p>Base &#8212; \u043f\u0440\u0435\u0434\u043d\u0430\u0437\u043d\u0430\u0447\u0435\u043d\u0430 \u0434\u043b\u044f \u0433\u0435\u043d\u0435\u0440\u0430\u0446\u0438\u0438 \u0441\u0438\u043d\u0442\u0435\u0442\u0438\u0447\u0435\u0441\u043a\u0438\u0445 \u0434\u0430\u043d\u043d\u044b\u0445.<\/p>\n<\/li>\n<li>\n<p>Insctuct &#8212; \u043f\u0440\u0435\u0434\u043d\u0430\u0437\u043d\u0430\u0447\u0435\u043d\u0430 \u0434\u043b\u044f \u0447\u0430\u0442\u0430 \u0438 \u0432\u044b\u043f\u043e\u043b\u043d\u0435\u043d\u0438\u044f \u0438\u043d\u0441\u0442\u0440\u0443\u043a\u0446\u0438\u0439<\/p>\n<\/li>\n<li>\n<p>Reward &#8212; base \u043c\u043e\u0434\u0435\u043b\u044c \u0441 \u0434\u043e\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u044b\u043c \u043b\u0438\u043d\u0435\u0439\u043d\u044b\u043c \u0441\u043b\u043e\u0435\u043c \u0434\u043b\u044f \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u0443\u044f \u043d\u043e\u0432\u044b\u0439 \u043f\u043e\u0434\u0445\u043e\u0434 reward.<\/p>\n<\/li>\n<\/ul>\n<p>\u041c\u043e\u0434\u0435\u043b\u044c \u0440\u0430\u0441\u043f\u0440\u043e\u0441\u0442\u0440\u0430\u043d\u044f\u0435\u0442\u0441\u044f \u043f\u043e\u0434 \u043b\u0438\u0446\u0435\u043d\u0437\u0438\u0435\u0439 NVIDIA Open Model License Agreement, \u0440\u0430\u0437\u0440\u0435\u0448\u0430\u044e\u0449\u0430\u044f \u043a\u043e\u043c\u043c\u0435\u0440\u0447\u0435\u0441\u043a\u043e\u0435 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u043d\u0438\u0435.<\/p>\n<p>\u0421\u0430\u043c\u0430 \u043c\u043e\u0434\u0435\u043b\u044c \u043c\u043e\u0436\u0435\u0442 \u0438 \u0432 \u0441\u0442\u0438\u0445\u0438, \u0442\u0430\u043a \u043a\u0430\u043a \u043e\u0431\u0443\u0447\u0430\u043b\u0430\u0441\u044c \u0438 \u043d\u0430 \u0440\u0443\u0441\u0441\u043a\u043e\u043c \u044f\u0437\u044b\u043a\u0435:<\/p>\n<figure class=\"full-width\"><\/figure>\n<p>\u041e\u043d\u043b\u0430\u0439\u043d \u0434\u0435\u043c\u043e: <a href=\"https:\/\/chat.lmsys.org\/\" rel=\"noopener noreferrer nofollow\">https:\/\/chat.lmsys.org\/<\/a> (\u0442\u0430\u043c \u0432\u044b\u0431\u0440\u0430\u0442\u044c Direct Chat)<\/p>\n<p>\u0412\u0435\u0441\u0430 \u043c\u043e\u0434\u0435\u043b\u0438: <a href=\"https:\/\/huggingface.co\/nvidia\/Nemotron-4-340B-Instruct\" rel=\"noopener noreferrer nofollow\">https:\/\/huggingface.co\/nvidia\/Nemotron-4-340B-Instruct<\/a><\/p>\n<p>safetensors: <a href=\"https:\/\/huggingface.co\/failspy\/Nemotron-4-340B-Instruct-SafeTensors\" rel=\"noopener noreferrer nofollow\">https:\/\/huggingface.co\/failspy\/Nemotron-4-340B-Instruct-SafeTensors<\/a><\/p>\n<p>\u0411\u043e\u043b\u044c\u0448\u0435 \u043f\u043e\u0434\u0440\u043e\u0431\u043d\u043e\u0441\u0442\u0435\u0439 \u0432 <a href=\"https:\/\/d1qx31qr3h6wln.cloudfront.net\/publications\/Nemotron_4_340B_8T_0.pdf\" rel=\"noopener noreferrer nofollow\">\u043f\u0440\u0435\u0441\u0441-\u0440\u0435\u043b\u0438\u0437\u0435<\/a>.<\/p>\n<p>\u0421\u0430\u043c\u0430\u044f \u0431\u043e\u043b\u044c\u0448\u0430\u044f \u0437\u0430\u0433\u0430\u0434\u043a\u0430 &#8212; \u0441\u043c\u043e\u0436\u0435\u0442 \u043b\u0438 1 \u0431\u0438\u0442\u043d\u043e\u0435 \u043a\u0432\u0430\u043d\u0442\u043e\u0432\u0430\u043d\u0438\u0435 gguf \u043f\u043e\u0437\u0432\u043e\u043b\u0438\u0442\u044c \u0437\u0430\u043f\u0443\u0441\u0442\u0438\u0442\u044c \u0435\u0451 \u043b\u043e\u043a\u0430\u043b\u044c\u043d\u043e, \u0438 \u0447\u0442\u043e \u0434\u0430\u0441\u0442 \u044d\u0442\u0430 \u043c\u043e\u0434\u0435\u043b\u044c \u0434\u043b\u044f \u0441\u043e\u043e\u0431\u0449\u0435\u0441\u0442\u0432\u0430 llm.<\/p>\n<\/p>\n<\/div>\n<\/div>\n<\/div>\n<p><!----><!----><\/div>\n<p><!----><!----><br \/> \u0441\u0441\u044b\u043b\u043a\u0430 \u043d\u0430 \u043e\u0440\u0438\u0433\u0438\u043d\u0430\u043b \u0441\u0442\u0430\u0442\u044c\u0438 <a href=\"https:\/\/habr.com\/ru\/articles\/822167\/\"> https:\/\/habr.com\/ru\/articles\/822167\/<\/a><br \/><\/br><\/br><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[],"tags":[],"class_list":["post-378995","post","type-post","status-publish","format-standard","hentry"],"_links":{"self":[{"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=\/wp\/v2\/posts\/378995","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=378995"}],"version-history":[{"count":0,"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=\/wp\/v2\/posts\/378995\/revisions"}],"wp:attachment":[{"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=378995"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=378995"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=378995"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}