{"id":1676,"date":"2024-04-26T02:40:10","date_gmt":"2024-04-26T02:40:10","guid":{"rendered":"https:\/\/tastycounter.net\/index.php\/2024\/04\/26\/apple-phat-hanh-nhieu-mo-hinh-ai-ma-nguon-mo-moi\/"},"modified":"2024-04-26T02:40:10","modified_gmt":"2024-04-26T02:40:10","slug":"apple-phat-hanh-nhieu-mo-hinh-ai-ma-nguon-mo-moi","status":"publish","type":"post","link":"https:\/\/tastycounter.net\/index.php\/2024\/04\/26\/apple-phat-hanh-nhieu-mo-hinh-ai-ma-nguon-mo-moi\/","title":{"rendered":"Apple ph\u00e1t h\u00e0nh nhi\u1ec1u m\u00f4 h\u00ecnh AI m\u00e3 ngu\u1ed3n m\u1edf m\u1edbi"},"content":{"rendered":"<\/p>\n
\n

Apple kh\u00f4ng chia s\u1ebb nhi\u1ec1u v\u1ec1 c\u00e1c k\u1ebf ho\u1ea1ch Generative AI c\u1ee7a m\u00ecnh. Gi\u1edd \u0111\u00e2y, v\u1edbi vi\u1ec7c ph\u00e1t h\u00e0nh m\u1ed9t d\u00f2ng m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef l\u1edbn m\u00e3 ngu\u1ed3n m\u1edf, c\u00f3 v\u1ebb nh\u01b0 g\u00e3 kh\u1ed5ng l\u1ed3 c\u00f4ng ngh\u1ec7 n\u00e0y \u0111ang mu\u1ed1n l\u00e0m cho AI ch\u1ea1y c\u1ee5c b\u1ed9 tr\u00ean c\u00e1c thi\u1ebft b\u1ecb Apple. LLM c\u1ee7a Apple, g\u1ecdi l\u00e0 OpenELM (Open-source Efficient Language Models), \u0111\u01b0\u1ee3c thi\u1ebft k\u1ebf \u0111\u1ec3 ch\u1ea1y tr\u00ean thi\u1ebft b\u1ecb thay v\u00ec tr\u00ean m\u00e1y ch\u1ee7 \u0111\u00e1m m\u00e2y. C\u00e1c LLM n\u00e0y c\u00f3 s\u1eb5n tr\u00ean Hugging Face Hub, m\u1ed9t n\u1ec1n t\u1ea3ng trung t\u00e2m \u0111\u1ec3 chia s\u1ebb code v\u00e0 b\u1ed9 d\u1eef li\u1ec7u AI.<\/p>\n

Trong qu\u00e1 tr\u00ecnh th\u1eed nghi\u1ec7m, Apple nh\u1eadn th\u1ea5y r\u1eb1ng OpenELM mang l\u1ea1i hi\u1ec7u su\u1ea5t t\u01b0\u01a1ng t\u1ef1 nh\u01b0 c\u00e1c m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef m\u1edf kh\u00e1c, nh\u01b0ng c\u00f3 \u00edt d\u1eef li\u1ec7u \u0111\u00e0o t\u1ea1o h\u01a1n.<\/p>\n

\"Apple
Apple ph\u00e1t h\u00e0nh 4 m\u00f4 h\u00ecnh AI ngu\u1ed3n m\u1edf m\u1edbi ch\u1ea1y tr\u00ean thi\u1ebft b\u1ecb<\/figcaption><\/figure>\n
<\/ins><\/div>\n

Nh\u01b0 \u0111\u01b0\u1ee3c tr\u00ecnh b\u00e0y, c\u00f3 t\u1ed5ng c\u1ed9ng 8 m\u00f4 h\u00ecnh OpenELM. 4 m\u00f4 h\u00ecnh \u0111\u00e3 \u0111\u01b0\u1ee3c \u0111\u00e0o t\u1ea1o tr\u01b0\u1edbc b\u1eb1ng th\u01b0 vi\u1ec7n CoreNet, trong khi 4 m\u00f4 h\u00ecnh c\u00f2n l\u1ea1i \u0111\u01b0\u1ee3c \u0111i\u1ec1u ch\u1ec9nh theo h\u01b0\u1edbng d\u1eabn. \u0110\u1ec3 c\u1ea3i thi\u1ec7n \u0111\u1ed9 ch\u00ednh x\u00e1c v\u00e0 hi\u1ec7u qu\u1ea3 t\u1ed5ng th\u1ec3, Apple s\u1eed d\u1ee5ng chi\u1ebfn l\u01b0\u1ee3c m\u1edf r\u1ed9ng quy m\u00f4 theo l\u1edbp trong c\u00e1c LLM ngu\u1ed3n m\u1edf n\u00e0y.<\/p>\n

\u201c\u0110\u1ec3 \u0111\u1ea1t \u0111\u01b0\u1ee3c m\u1ee5c ti\u00eau n\u00e0y, ch\u00fang t\u00f4i ph\u00e1t h\u00e0nh OpenELM, m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef m\u1edf ti\u00ean ti\u1ebfn nh\u1ea5t. OpenELM s\u1eed d\u1ee5ng chi\u1ebfn l\u01b0\u1ee3c m\u1edf r\u1ed9ng quy m\u00f4 theo l\u1edbp \u0111\u1ec3 ph\u00e2n b\u1ed5 hi\u1ec7u qu\u1ea3 c\u00e1c tham s\u1ed1 trong m\u1ed7i l\u1edbp c\u1ee7a m\u00f4 h\u00ecnh transformer, gi\u00fap n\u00e2ng cao \u0111\u1ed9 ch\u00ednh x\u00e1c. V\u00ed d\u1ee5, v\u1edbi kho\u1ea3ng m\u1ed9t t\u1ef7 tham s\u1ed1, OpenELM th\u1ec3 hi\u1ec7n \u0111\u1ed9 ch\u00ednh x\u00e1c \u0111\u01b0\u1ee3c c\u1ea3i thi\u1ec7n 2,36% so v\u1edbi OLMo trong khi y\u00eau c\u1ea7u token \u0111\u00e0o t\u1ea1o tr\u01b0\u1edbc \u00edt h\u01a1n 2 l\u1ea7n\u201c <\/em>– theo Apple.<\/p>\n

Apple kh\u00f4ng ch\u1ec9 cung c\u1ea5p m\u00f4 h\u00ecnh \u0111\u01b0\u1ee3c \u0111\u00e0o t\u1ea1o m\u1edbi nh\u1ea5t. \u0110\u00fang h\u01a1n, n\u00f3 c\u00f2n cung c\u1ea5p code, nh\u1eadt k\u00fd hu\u1ea5n luy\u1ec7n v\u00e0 nhi\u1ec1u phi\u00ean b\u1ea3n c\u1ee7a m\u00f4 h\u00ecnh. C\u00e1c nh\u00e0 nghi\u00ean c\u1ee9u d\u1ef1 \u00e1n l\u1ea1c quan cho r\u1eb1ng n\u00f3 s\u1ebd \u0111\u1ea9y nhanh nh\u1eefng ti\u1ebfn b\u1ed9 v\u00e0 mang l\u1ea1i \u201ck\u1ebft qu\u1ea3 \u0111\u00e1ng tin c\u1eady\u201d<\/em> trong l\u0129nh v\u1ef1c AI s\u1eed d\u1ee5ng ng\u00f4n ng\u1eef t\u1ef1 nhi\u00ean.<\/p>\n

\u201cKh\u00e1c v\u1edbi c\u00e1c ph\u01b0\u01a1ng ph\u00e1p tr\u01b0\u1edbc \u0111\u00e2y ch\u1ec9 cung c\u1ea5p tr\u1ecdng s\u1ed1 m\u00f4 h\u00ecnh v\u00e0 code suy lu\u1eadn c\u0169ng nh\u01b0 \u0111\u00e0o t\u1ea1o tr\u01b0\u1edbc tr\u00ean c\u00e1c t\u1eadp d\u1eef li\u1ec7u ri\u00eang t\u01b0, b\u1ea3n ph\u00e1t h\u00e0nh c\u1ee7a ch\u00fang t\u00f4i bao g\u1ed3m framework ho\u00e0n ch\u1ec9nh \u0111\u1ec3 \u0111\u00e0o t\u1ea1o v\u00e0 \u0111\u00e1nh gi\u00e1 m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef tr\u00ean c\u00e1c t\u1eadp d\u1eef li\u1ec7u c\u00f3 s\u1eb5n c\u00f4ng khai, bao g\u1ed3m nh\u1eadt k\u00fd \u0111\u00e0o t\u1ea1o, nhi\u1ec1u checkpoint v\u00e0 c\u1ea5u h\u00ecnh \u0111\u01b0\u1ee3c \u0111\u00e0o t\u1ea1o tr\u01b0\u1edbc. Ch\u00fang t\u00f4i c\u0169ng ph\u00e1t h\u00e0nh code chuy\u1ec3n \u0111\u1ed5i m\u00f4 h\u00ecnh sang th\u01b0 vi\u1ec7n MLX \u0111\u1ec3 suy lu\u1eadn v\u00e0 tinh ch\u1ec9nh tr\u00ean c\u00e1c thi\u1ebft b\u1ecb Apple. B\u1ea3n ph\u00e1t h\u00e0nh to\u00e0n di\u1ec7n n\u00e0y nh\u1eb1m m\u1ee5c \u0111\u00edch trao quy\u1ec1n v\u00e0 c\u1ee7ng c\u1ed1 c\u1ed9ng \u0111\u1ed3ng nghi\u00ean c\u1ee9u m\u1edf, m\u1edf \u0111\u01b0\u1eddng cho nh\u1eefng n\u1ed7 l\u1ef1c nghi\u00ean c\u1ee9u trong t\u01b0\u01a1ng lai\u201d <\/em>– theo Apple.<\/p>\n

Apple n\u00f3i th\u00eam r\u1eb1ng vi\u1ec7c ph\u00e1t h\u00e0nh c\u00e1c m\u00f4 h\u00ecnh OpenELM s\u1ebd \u201ctrao quy\u1ec1n v\u00e0 l\u00e0m phong ph\u00fa th\u00eam c\u1ed9ng \u0111\u1ed3ng nghi\u00ean c\u1ee9u m\u1edf\u201d<\/em> v\u1edbi c\u00e1c m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef hi\u1ec7n \u0111\u1ea1i. Nh\u1eefng m\u00f4 h\u00ecnh ngu\u1ed3n m\u1edf cho ph\u00e9p c\u00e1c nh\u00e0 nghi\u00ean c\u1ee9u kh\u00e1m ph\u00e1 nh\u1eefng r\u1ee7i ro, d\u1eef li\u1ec7u v\u00e0 \u0111i\u1ec3m y\u1ebfu ti\u1ec1m \u1ea9n v\u1ed1n c\u00f3 trong c\u00e1c m\u00f4 h\u00ecnh. C\u00e1c m\u00f4 h\u00ecnh \u0111\u00e3 s\u1eb5n s\u00e0ng \u0111\u1ec3 nh\u00e0 ph\u00e1t tri\u1ec3n s\u1eed d\u1ee5ng. H\u1ecd c\u00f3 th\u1ec3 s\u1eed d\u1ee5ng nguy\u00ean tr\u1ea1ng LLM ngu\u1ed3n m\u1edf n\u00e0y ho\u1eb7c th\u1ef1c hi\u1ec7n nh\u1eefng s\u1eeda \u0111\u1ed5i c\u1ea7n thi\u1ebft.<\/p>\n

<\/ins><\/div>\n

Tr\u1edf l\u1ea1i v\u00e0o th\u00e1ng 2, CEO Tim Cook c\u1ee7a Apple \u0111\u00e3 ti\u1ebft l\u1ed9 r\u1eb1ng c\u00e1c t\u00ednh n\u0103ng Generative AI<\/a> s\u1ebd \u0111\u1ebfn v\u1edbi nh\u1eefng thi\u1ebft b\u1ecb c\u1ee7a Apple v\u00e0o cu\u1ed1i n\u0103m nay. M\u1ed9t th\u1eddi gian sau \u0111\u00f3, \u00f4ng nh\u1eafc l\u1ea1i r\u1eb1ng c\u00f4ng ty \u0111ang n\u1ed7 l\u1ef1c \u0111\u1ec3 mang l\u1ea1i nh\u1eefng tr\u1ea3i nghi\u1ec7m AI \u0111\u1ed9t ph\u00e1.<\/p>\n

Tr\u01b0\u1edbc \u0111\u00e2y, Apple \u0111\u00e3 ph\u00e1t h\u00e0nh m\u1ed9t s\u1ed1 m\u00f4 h\u00ecnh AI kh\u00e1c. Th\u1eadt ti\u1ebfc l\u00e0 h\u00e3ng n\u00e0y v\u1eabn ch\u01b0a mang kh\u1ea3 n\u0103ng AI v\u00e0o c\u00e1c thi\u1ebft b\u1ecb c\u1ee7a m\u00ecnh. Tuy nhi\u00ean, iOS 18<\/a> s\u1eafp t\u1edbi d\u1ef1 ki\u1ebfn \u200b\u200bs\u1ebd bao g\u1ed3m m\u1ed9t b\u1ed9 t\u00ednh n\u0103ng AI m\u1edbi v\u00e0 b\u1ea3n ph\u00e1t h\u00e0nh OpenELM c\u00f3 th\u1ec3 l\u00e0 ph\u1ea7n chu\u1ea9n b\u1ecb h\u1eadu tr\u01b0\u1eddng m\u1edbi nh\u1ea5t c\u1ee7a Apple.<\/p>\n

G\u1ea7n \u0111\u00e2y, Mark Gurman c\u0169ng b\u00e1o c\u00e1o r\u1eb1ng c\u00e1c t\u00ednh n\u0103ng AI c\u1ee7a iOS 18 s\u1ebd ch\u1ee7 y\u1ebfu \u0111\u01b0\u1ee3c h\u1ed7 tr\u1ee3 b\u1edfi m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef l\u1edbn tr\u00ean thi\u1ebft b\u1ecb nh\u1eb1m mang l\u1ea1i l\u1ee3i \u00edch v\u1ec1 quy\u1ec1n ri\u00eang t\u01b0 v\u00e0 t\u1ed1c \u0111\u1ed9. Ch\u00fang ta s\u1ebd bi\u1ebft m\u1ecdi th\u1ee9 khi Apple c\u00f4ng b\u1ed1 iOS 18 v\u00e0 c\u00e1c b\u1ea3n n\u00e2ng c\u1ea5p ph\u1ea7n m\u1ec1m kh\u00e1c t\u1ea1i WWDC v\u00e0o ng\u00e0y 10 th\u00e1ng 6 t\u1edbi \u0111\u00e2y.<\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

Apple kh\u00f4ng chia s\u1ebb nhi\u1ec1u v\u1ec1 c\u00e1c k\u1ebf ho\u1ea1ch Generative AI c\u1ee7a m\u00ecnh. Gi\u1edd \u0111\u00e2y, v\u1edbi vi\u1ec7c ph\u00e1t h\u00e0nh m\u1ed9t d\u00f2ng m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef l\u1edbn m\u00e3 ngu\u1ed3n m\u1edf, c\u00f3 v\u1ebb nh\u01b0 g\u00e3 kh\u1ed5ng l\u1ed3 c\u00f4ng ngh\u1ec7 n\u00e0y \u0111ang mu\u1ed1n l\u00e0m cho AI ch\u1ea1y c\u1ee5c b\u1ed9 tr\u00ean c\u00e1c thi\u1ebft b\u1ecb Apple. LLM c\u1ee7a Apple, g\u1ecdi […]<\/p>\n","protected":false},"author":1,"featured_media":1677,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-1676","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-khong-phan-loai"],"_links":{"self":[{"href":"https:\/\/tastycounter.net\/index.php\/wp-json\/wp\/v2\/posts\/1676","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/tastycounter.net\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/tastycounter.net\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/tastycounter.net\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/tastycounter.net\/index.php\/wp-json\/wp\/v2\/comments?post=1676"}],"version-history":[{"count":0,"href":"https:\/\/tastycounter.net\/index.php\/wp-json\/wp\/v2\/posts\/1676\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/tastycounter.net\/index.php\/wp-json\/wp\/v2\/media\/1677"}],"wp:attachment":[{"href":"https:\/\/tastycounter.net\/index.php\/wp-json\/wp\/v2\/media?parent=1676"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/tastycounter.net\/index.php\/wp-json\/wp\/v2\/categories?post=1676"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/tastycounter.net\/index.php\/wp-json\/wp\/v2\/tags?post=1676"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}