{"id":104,"date":"2023-11-15T15:23:55","date_gmt":"2023-11-15T08:23:55","guid":{"rendered":"https:\/\/xuhuongai.com\/?p=104"},"modified":"2023-11-18T12:19:53","modified_gmt":"2023-11-18T05:19:53","slug":"doi-ngu-hugging-face-chi-voi-hai-nguoi-dang-phat-trien-mo-hinh-giong-voi-chatgpt","status":"publish","type":"post","link":"https:\/\/xuhuongai.com\/?p=104","title":{"rendered":"\u0110\u1ed9i ng\u0169 ch\u1ec9 v\u1edbi hai ng\u01b0\u1eddi c\u1ee7a Hugging Face \u0111ang ph\u00e1t tri\u1ec3n m\u00f4 h\u00ecnh gi\u1ed1ng v\u1edbi ChatGPT"},"content":{"rendered":"\n
\n

B\u1ea3n tin \u0111\u01b0\u1ee3c d\u1ecbch v\u00e0 t\u00f3m t\u1eaft b\u1edfi n\u1ec1n t\u1ea3ng t\u1ea1o tr\u1ee3 l\u00fd AI – KamiMind<\/a>.<\/p>\nNgu\u1ed3n: Kyle Wiggers, “Hugging Face has a two-person team developing ChatGPT-like AI models<\/a>“, TechCrunch, 09\/11\/2023.<\/cite><\/blockquote>\n\n\n

\n
\"\"
\u1ea2nh minh h\u1ecda. Ngu\u1ed3n: <\/strong>Hugging Face<\/figcaption><\/figure>\n<\/div>\n\n\n

C\u00f4ng ty Hugging Face, m\u1ed9t c\u00f4ng ty kh\u1edfi nghi\u1ec7p tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o, \u0111\u00e3 ph\u00e1t tri\u1ec3n m\u1ed9t lo\u1ea1t c\u00f4ng c\u1ee5 \u0111\u1ec3 h\u1ed7 tr\u1ee3 ph\u00e1t tri\u1ec3n v\u00e0 l\u01b0u tr\u1eef d\u1eef li\u1ec7u khoa h\u1ecdc. C\u00f4ng ty n\u00e0y cung c\u1ea5p m\u1ed9t c\u1ed5ng th\u00f4ng tin gi\u1ed1ng GitHub \u0111\u1ec3 l\u01b0u tr\u1eef m\u00e3 AI, m\u00f4 h\u00ecnh v\u00e0 b\u1ed9 d\u1eef li\u1ec7u, c\u00f9ng v\u1edbi m\u1ed9t b\u1ea3ng \u0111i\u1ec1u khi\u1ec3n web \u0111\u1ec3 tr\u00ecnh di\u1ec5n c\u00e1c \u1ee9ng d\u1ee5ng tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o. M\u1ed9t trong nh\u1eefng c\u00f4ng c\u1ee5 \u1ea5n t\u01b0\u1ee3ng nh\u1ea5t c\u1ee7a Hugging Face l\u00e0 H4, m\u1ed9t \u0111\u1ed9i ng\u0169 hai ng\u01b0\u1eddi chuy\u00ean ph\u00e1t tri\u1ec3n c\u00e1c c\u00f4ng c\u1ee5 v\u00e0 “c\u00f4ng th\u1ee9c” \u0111\u1ec3 x\u00e2y d\u1ef1ng c\u00e1c chatbot d\u1ef1a tr\u00ean tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o. \u0110\u1ed9i ng\u0169 n\u00e0y \u0111\u00e3 t\u1ea1o ra c\u00e1c m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef m\u00e3 ngu\u1ed3n m\u1edf nh\u01b0 Zephyr-7B-\u03b1 v\u00e0 Falcon-40B, v\u00e0 s\u1eed d\u1ee5ng m\u1ed9t c\u1ee5m m\u00e1y ch\u1ee7 \u0111\u1eb7c bi\u1ec7t \u0111\u1ec3 hu\u1ea5n luy\u1ec7n m\u00f4 h\u00ecnh. H4 kh\u00f4ng tr\u1ef1c ti\u1ebfp th\u01b0\u01a1ng m\u1ea1i h\u00f3a c\u00f4ng c\u1ee5 c\u1ee7a m\u00ecnh, nh\u01b0ng \u0111\u00f3ng g\u00f3p v\u00e0o Ch\u01b0\u01a1ng tr\u00ecnh T\u0103ng t\u1ed1c Chuy\u00ean gia c\u1ee7a Hugging Face \u0111\u1ec3 x\u00e2y d\u1ef1ng c\u00e1c gi\u1ea3i ph\u00e1p tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o t\u00f9y ch\u1ec9nh. \u0110\u1ed9i ng\u0169 c\u0169ng h\u1ee3p t\u00e1c v\u1edbi c\u1ed9ng \u0111\u1ed3ng m\u00e3 ngu\u1ed3n m\u1edf b\u1eb1ng c\u00e1ch ph\u00e1t h\u00e0nh m\u00e3 hu\u1ea5n luy\u1ec7n v\u00e0 b\u1ed9 d\u1eef li\u1ec7u li\u00ean quan \u0111\u1ebfn m\u00f4 h\u00ecnh chat.<\/p>\n\n\n\n

B\u1ea3n t\u00f3m t\u1eaft ti\u1ebfng Anh<\/summary>\n

AI startup Hugging Face offers a range of tools for data science hosting and development, including a GitHub-like platform for AI code and web dashboards for AI demos. One of their impressive tools is H4, a two-person team formed to develop tools and “recipes” for building AI-powered chatbots similar to ChatGPT. H4 has focused on alignment research, teaching language models how to behave based on feedback. They have released open source models like Zephyr-7B-\u03b1 and modified Falcon-40B for better natural language responses. H4 relies on a dedicated cluster of Nvidia GPUs for model training. Although H4 doesn’t directly monetize its tools, they contribute to Hugging Face’s Expert Acceleration Program. H4 aims to empower the open AI community by releasing training code and datasets. They collaborate with external groups and value contributions from the community.<\/p>\n<\/details>\n\n\n\n

B\u1ea3n d\u1ecbch Anh – Vi\u1ec7t<\/summary>\n

C\u00f4ng ty kh\u1edfi nghi\u1ec7p tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o Hugging Face cung c\u1ea5p m\u1ed9t lo\u1ea1t c\u00e1c c\u00f4ng c\u1ee5 ph\u00e1t tri\u1ec3n v\u00e0 l\u01b0u tr\u1eef khoa h\u1ecdc d\u1eef li\u1ec7u, bao g\u1ed3m m\u1ed9t c\u1ed5ng th\u00f4ng tin gi\u1ed1ng GitHub \u0111\u1ec3 l\u01b0u tr\u1eef m\u00e3 AI, m\u00f4 h\u00ecnh v\u00e0 b\u1ed9 d\u1eef li\u1ec7u, c\u0169ng nh\u01b0 b\u1ea3ng \u0111i\u1ec1u khi\u1ec3n web \u0111\u1ec3 tr\u00ecnh di\u1ec5n c\u00e1c \u1ee9ng d\u1ee5ng d\u1ef1a tr\u00ean tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o.<\/p>\n\n\n\n

Nh\u01b0ng m\u1ed9t s\u1ed1 c\u00f4ng c\u1ee5 \u1ea5n t\u01b0\u1ee3ng v\u00e0 c\u00f3 kh\u1ea3 n\u0103ng cao nh\u1ea5t c\u1ee7a Hugging Face hi\u1ec7n nay \u0111\u1ebfn t\u1eeb m\u1ed9t \u0111\u1ed9i ng\u0169 hai ng\u01b0\u1eddi \u0111\u01b0\u1ee3c th\u00e0nh l\u1eadp v\u00e0o th\u00e1ng 1.<\/p>\n\n\n\n

H4, c\u00f2n \u0111\u01b0\u1ee3c g\u1ecdi l\u00e0 “H4” vi\u1ebft t\u1eaft c\u1ee7a “h\u1eefu \u00edch, ch\u00e2n th\u00e0nh, v\u00f4 h\u1ea1i v\u00e0 \u00e2u y\u1ebfm”, nh\u1eb1m ph\u00e1t tri\u1ec3n c\u00e1c c\u00f4ng c\u1ee5 v\u00e0 “c\u00f4ng th\u1ee9c” \u0111\u1ec3 cho ph\u00e9p c\u1ed9ng \u0111\u1ed3ng tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o x\u00e2y d\u1ef1ng c\u00e1c chatbot d\u1ef1a tr\u00ean tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o theo h\u01b0\u1edbng ChatGPT. Vi\u1ec7c ra m\u1eaft ChatGPT \u0111\u00e3 l\u00e0 nguy\u00ean nh\u00e2n d\u1eabn \u0111\u1ebfn vi\u1ec7c th\u00e0nh l\u1eadp H4, theo Lewis Tunstall, m\u1ed9t k\u1ef9 s\u01b0 h\u1ecdc m\u00e1y t\u1ea1i Hugging Face v\u00e0 m\u1ed9t trong hai th\u00e0nh vi\u00ean c\u1ee7a H4.<\/p>\n\n\n\n

“Khi ChatGPT \u0111\u01b0\u1ee3c OpenAI ra m\u1eaft v\u00e0o cu\u1ed1i n\u0103m 2022, ch\u00fang t\u00f4i \u0111\u00e3 b\u1eaft \u0111\u1ea7u th\u1ea3o lu\u1eadn v\u1ec1 nh\u1eefng g\u00ec c\u00f3 th\u1ec3 c\u1ea7n \u0111\u1ec3 sao ch\u00e9p kh\u1ea3 n\u0103ng c\u1ee7a n\u00f3 b\u1eb1ng c\u00e1c th\u01b0 vi\u1ec7n m\u00e3 ngu\u1ed3n m\u1edf v\u00e0 c\u00e1c m\u00f4 h\u00ecnh,” Tunstall cho bi\u1ebft trong cu\u1ed9c ph\u1ecfng v\u1ea5n qua email v\u1edbi TechCrunch. “Nghi\u00ean c\u1ee9u ch\u00ednh c\u1ee7a H4 t\u1eadp trung v\u00e0o vi\u1ec7c \u0111\u1ed3ng b\u1ed9, m\u00e0 bao g\u1ed3m d\u1ea1y cho LLMs c\u00e1ch c\u01b0 x\u1eed theo ph\u1ea3n h\u1ed3i t\u1eeb con ng\u01b0\u1eddi (ho\u1eb7c th\u1eadm ch\u00ed l\u00e0 c\u00e1c tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o kh\u00e1c).”<\/p>\n\n\n\n

H4 \u0111\u1ee9ng sau m\u1ed9t s\u1ed1 l\u01b0\u1ee3ng ng\u00e0y c\u00e0ng t\u0103ng c\u00e1c m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef l\u1edbn m\u00e3 ngu\u1ed3n m\u1edf, bao g\u1ed3m Zephyr-7B-\u03b1, m\u1ed9t phi\u00ean b\u1ea3n tinh ch\u1ec9nh, t\u1eadp trung v\u00e0o chat, c\u1ee7a m\u00f4 h\u00ecnh Mistral 7B do c\u00f4ng ty kh\u1edfi nghi\u1ec7p tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o Ph\u00e1p Mistral m\u1edbi \u0111\u00e2y ra m\u1eaft. H4 c\u0169ng \u0111\u00e3 \u0111\u00e1nh nh\u00e1nh Falcon-40B, m\u1ed9t m\u00f4 h\u00ecnh t\u1eeb Vi\u1ec7n \u0110\u1ed5i m\u1edbi C\u00f4ng ngh\u1ec7 t\u1ea1i Abu Dhabi – s\u1eeda \u0111\u1ed5i m\u00f4 h\u00ecnh \u0111\u1ec3 \u0111\u00e1p \u1ee9ng m\u1ed9t c\u00e1ch h\u1eefu \u00edch h\u01a1n c\u00e1c y\u00eau c\u1ea7u b\u1eb1ng ng\u00f4n ng\u1eef t\u1ef1 nhi\u00ean.<\/p>\n\n\n\n

\u0110\u1ec3 hu\u1ea5n luy\u1ec7n c\u00e1c m\u00f4 h\u00ecnh c\u1ee7a m\u00ecnh, H4 – gi\u1ed1ng nh\u01b0 c\u00e1c nh\u00f3m nghi\u00ean c\u1ee9u kh\u00e1c c\u1ee7a Hugging Face – d\u1ef1a v\u00e0o m\u1ed9t c\u1ee5m m\u00e1y ch\u1ee7 \u0111\u1eb7c bi\u1ec7t v\u1edbi h\u01a1n 1.000 GPU Nvidia A100. Tunstall v\u00e0 \u0111\u1ed3ng nghi\u1ec7p H4 kh\u00e1c, Ed Beeching, \u0111\u1eb7t t\u1ea1i ch\u1ed7 \u1edf ch\u00e2u \u00c2u, nh\u01b0ng nh\u1eadn \u0111\u01b0\u1ee3c s\u1ef1 h\u1ed7 tr\u1ee3 t\u1eeb m\u1ed9t s\u1ed1 nh\u00f3m Hugging Face n\u1ed9i b\u1ed9, bao g\u1ed3m nh\u00f3m ki\u1ec3m tra v\u00e0 \u0111\u00e1nh gi\u00e1 m\u00f4 h\u00ecnh.<\/p>\n\n\n\n

“K\u00edch th\u01b0\u1edbc nh\u1ecf c\u1ee7a H4 l\u00e0 m\u1ed9t l\u1ef1a ch\u1ecdn c\u1ed1 \u00fd, v\u00ec n\u00f3 cho ph\u00e9p ch\u00fang t\u00f4i linh ho\u1ea1t h\u01a1n v\u00e0 th\u00edch \u1ee9ng v\u1edbi m\u1ed9t c\u1ea3nh nghi\u00ean c\u1ee9u thay \u0111\u1ed5i li\u00ean t\u1ee5c,” Beeching cho bi\u1ebft trong cu\u1ed9c ph\u1ecfng v\u1ea5n qua email v\u1edbi TechCrunch. “Ch\u00fang t\u00f4i c\u0169ng c\u00f3 m\u1ed9t s\u1ed1 h\u1ee3p t\u00e1c b\u00ean ngo\u00e0i v\u1edbi c\u00e1c nh\u00f3m nh\u01b0 LMSYS v\u00e0 LlamaIndex, ch\u00fang t\u00f4i h\u1ee3p t\u00e1c \u0111\u1ec3 ph\u00e1t h\u00e0nh chung.”<\/p>\n\n\n\n

G\u1ea7n \u0111\u00e2y, H4 \u0111\u00e3 nghi\u00ean c\u1ee9u c\u00e1c k\u1ef9 thu\u1eadt \u0111\u1ed3ng b\u1ed9 kh\u00e1c nhau v\u00e0 x\u00e2y d\u1ef1ng c\u00e1c c\u00f4ng c\u1ee5 \u0111\u1ec3 ki\u1ec3m tra hi\u1ec7u qu\u1ea3 c\u1ee7a c\u00e1c k\u1ef9 thu\u1eadt \u0111\u01b0\u1ee3c \u0111\u1ec1 xu\u1ea5t b\u1edfi c\u1ed9ng \u0111\u1ed3ng v\u00e0 ng\u00e0nh c\u00f4ng nghi\u1ec7p. \u0110\u1ed9i ng\u0169 n\u00e0y \u0111\u00e3 ph\u00e1t h\u00e0nh m\u1ed9t cu\u1ed1n s\u1ed5 tay ch\u1ee9a to\u00e0n b\u1ed9 m\u00e3 ngu\u1ed3n v\u00e0 b\u1ed9 d\u1eef li\u1ec7u m\u00e0 h\u1ecd s\u1eed d\u1ee5ng \u0111\u1ec3 x\u00e2y d\u1ef1ng Zephyr, v\u00e0 H4 d\u1ef1 \u0111\u1ecbnh c\u1eadp nh\u1eadt s\u1ed5 tay v\u1edbi m\u00e3 t\u1eeb c\u00e1c m\u00f4 h\u00ecnh tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o t\u01b0\u01a1ng lai c\u1ee7a h\u1ecd khi ch\u00fang \u0111\u01b0\u1ee3c ph\u00e1t h\u00e0nh.<\/p>\n\n\n\n

T\u00f4i h\u1ecfi li\u1ec7u H4 c\u00f3 \u00e1p l\u1ef1c t\u1eeb c\u00e1c nh\u00e2n vi\u00ean c\u1ea5p cao c\u1ee7a Hugging Face \u0111\u1ec3 th\u01b0\u01a1ng m\u1ea1i h\u00f3a c\u00f4ng vi\u1ec7c c\u1ee7a h\u1ecd. Sau t\u1ea5t c\u1ea3, c\u00f4ng ty \u0111\u00e3 thu v\u1ec1 h\u00e0ng tr\u0103m tri\u1ec7u \u0111\u00f4 la t\u1eeb m\u1ed9t nh\u00f3m nh\u00e0 \u0111\u1ea7u t\u01b0 c\u00f3 ti\u1ebfng bao g\u1ed3m Salesforce, IBM, AMD, Google, Amazon Intel v\u00e0 Nvidia. V\u00f2ng huy \u0111\u1ed9ng v\u1ed1n cu\u1ed1i c\u00f9ng c\u1ee7a Hugging Face \u0111\u00e3 \u0111\u1ecbnh gi\u00e1 c\u00f4ng ty n\u00e0y \u1edf m\u1ee9c 4,5 t\u1ef7 \u0111\u00f4 la, g\u1ea5p \u0111\u00f4i so v\u1edbi doanh thu h\u00e0ng n\u0103m c\u1ee7a c\u00f4ng ty.<\/p>\n\n\n\n

Tunstall cho bi\u1ebft H4 kh\u00f4ng tr\u1ef1c ti\u1ebfp th\u01b0\u01a1ng m\u1ea1i h\u00f3a c\u00f4ng c\u1ee5 c\u1ee7a m\u00ecnh. Nh\u01b0ng \u00f4ng th\u1eeba nh\u1eadn r\u1eb1ng c\u00e1c c\u00f4ng c\u1ee5 n\u00e0y \u0111\u00f3ng g\u00f3p v\u00e0o Ch\u01b0\u01a1ng tr\u00ecnh T\u0103ng t\u1ed1c Chuy\u00ean gia c\u1ee7a Hugging Face, m\u1ed9t d\u1ecbch v\u1ee5 t\u1eadp trung v\u00e0o doanh nghi\u1ec7p, cung c\u1ea5p h\u01b0\u1edbng d\u1eabn t\u1eeb c\u00e1c nh\u00f3m Hugging Face \u0111\u1ec3 x\u00e2y d\u1ef1ng c\u00e1c gi\u1ea3i ph\u00e1p tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o t\u00f9y ch\u1ec9nh.<\/p>\n\n\n\n

Khi \u0111\u01b0\u1ee3c h\u1ecfi xem H4 c\u00f3 xem m\u00ecnh l\u00e0 \u0111\u1ed1i th\u1ee7 c\u1ee7a c\u00e1c s\u00e1ng ki\u1ebfn tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o m\u00e3 ngu\u1ed3n m\u1edf kh\u00e1c, nh\u01b0 EleutherAI v\u00e0 LAION, Beeching cho bi\u1ebft \u0111\u00f3 kh\u00f4ng ph\u1ea3i l\u00e0 m\u1ee5c ti\u00eau c\u1ee7a H4. Thay v\u00e0o \u0111\u00f3, \u00f4ng n\u00f3i, \u00fd \u0111\u1ecbnh l\u00e0 “t\u1ea1o \u0111i\u1ec1u ki\u1ec7n cho” c\u1ed9ng \u0111\u1ed3ng tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o m\u00e3 ngu\u1ed3n m\u1edf b\u1eb1ng c\u00e1ch ph\u00e1t h\u00e0nh m\u00e3 hu\u1ea5n luy\u1ec7n v\u00e0 b\u1ed9 d\u1eef li\u1ec7u li\u00ean quan \u0111\u1ebfn c\u00e1c m\u00f4 h\u00ecnh chat c\u1ee7a H4.<\/p>\n\n\n\n

“C\u00f4ng vi\u1ec7c c\u1ee7a ch\u00fang t\u00f4i kh\u00f4ng th\u1ec3 thi\u1ebfu nh\u1eefng \u0111\u00f3ng g\u00f3p c\u1ee7a c\u1ed9ng \u0111\u1ed3ng,” Beeching n\u00f3i.<\/p>\n<\/details>\n","protected":false},"excerpt":{"rendered":"

M\u1ed9t trong nh\u1eefng c\u00f4ng c\u1ee5 \u1ea5n t\u01b0\u1ee3ng nh\u1ea5t c\u1ee7a Hugging Face l\u00e0 H4, m\u1ed9t \u0111\u1ed9i ng\u0169 hai ng\u01b0\u1eddi chuy\u00ean ph\u00e1t tri\u1ec3n c\u00e1c c\u00f4ng c\u1ee5 v\u00e0 “c\u00f4ng th\u1ee9c” \u0111\u1ec3 x\u00e2y d\u1ef1ng c\u00e1c chatbot d\u1ef1a tr\u00ean tr\u00ed tu\u1ec7 nh\u00e2n t\u1ea1o.<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5],"tags":[21,22,20],"class_list":["post-104","post","type-post","status-publish","format-standard","hentry","category-ai-news","tag-chatgpt","tag-cong-dong-tri-tue-nhan-tao","tag-hugging-face"],"_links":{"self":[{"href":"https:\/\/xuhuongai.com\/index.php?rest_route=\/wp\/v2\/posts\/104","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/xuhuongai.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/xuhuongai.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/xuhuongai.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/xuhuongai.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=104"}],"version-history":[{"count":6,"href":"https:\/\/xuhuongai.com\/index.php?rest_route=\/wp\/v2\/posts\/104\/revisions"}],"predecessor-version":[{"id":446,"href":"https:\/\/xuhuongai.com\/index.php?rest_route=\/wp\/v2\/posts\/104\/revisions\/446"}],"wp:attachment":[{"href":"https:\/\/xuhuongai.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=104"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/xuhuongai.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=104"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/xuhuongai.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=104"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}