InoteveraGaidhi rinotevera
Musanganiswa-we-Agents Aggregation
Mutauro AI
Mutauro AI GUIDE
Musanganiswa Wezvakadzama
Musanganiswa Wezvakadzama (MoD) inoita kuti transformer ipedze huwandu hwakasiyana hwemakombuta pane akasiyana tokeni, ichiendesa chete 'akakosha' tokeni kuburikidza nechero layer inorema computation.
Teerera kugwaro iriAI-yakarondedzerwa · ~2 min
Pa peji ino2 min verenga
Pfupiso
It cuts the cost of processing easy tokens while keeping a fixed, predictable compute budget.
Kudzika Kwakadzika
Mashandurirwo akajairwa anoisa yese layer kune ese tokeni, nyangwe netudiki semapu. Musanganiswa Wezvakadzama, zvakaunzwa ne Google DeepMind muna 2024, inowedzera diki router pabhuroko rega rega inosarudza yakagadziriswa top-k chidimbu chetokens kuti iite iyo yakazara self-attention uye MLP computation; vamwe vose vanosvetuka block kuburikidza neasara yekubatanidza. Nekuti k chete ma tokeni anogadziriswa pachikamu, iyo yakazara compute (FLOPs) inovharwa uye inozivikanwa pachine nguva, kusiyana nekare nzira dzakadzama-dzakadzama dzakasiyana zvisingafungidzirwe. Izvi zvinoita kuti batching uye kushandiswa kwehardware kushande. MoD-yakadzidziswa modhi inogona kuenderana neiyo yekutanga shanduko yemhando ichishandisa mashoma maFLOPs pakupfuura kwemberi, kana kusvika pamhando yepamusoro pamakomputa mamwe chete, uye pfungwa yacho inoumba yakasikwa neMusanganiswa-we-Nyanzvi kuti ipe 'MoDE' modhi iyo nzira pakudzika nehupamhi.
Kuitwa muNyika yeChokwadi
Kudzikisa maFLOP anodiwa kugadzirisa magwaro marefu nekusvetuka yakadzika computation pane mafirita tokeni
Kudzidzira modhi inoenderana nemhando yekutanga pane yakaderera compute, kudzikisa mutengo wesevhisi
Kubatanidza neMusanganiswa-we-Nyanzvi (MoDE) kuenda kune ese ari maviri kudzika uye sarudzo yehunyanzvi
Kuchengeta fungidziro, yakamisikidzwa latency pachiratidzo nekuti per-layer compute bhajeti inogadziriswa pachine nguva
Ramangwana Remusanganiswa Wezvakadzama
Conditional computation ndiyo huru lever yekubudirira seyero yemhando, uye MoD ndiyo yekutanga, yakachena muenzaniso. Tarisira kubatanidzwa kwakadzama neMusanganiswa-we-Nyanzvi (kutenderera pane zvese zvakadzika uye nyanzvi), mabhajeti anochinja anoderera kuti apinze zviri nyore, uye ma router akadzidza anoona zviri nani kuti ndeapi ma tokens anoda zvakadzama kugadzirisa. Sezvo mutengo wekufungidzira uchitonga hupfumi hwekutumirwa, matekiniki anoita kuti mamodheru 'afunge zvakanyanya' pazvinenge zvichidikanwa chete, uku uchichengeta kurongeka kwekufungidzira, anogona kuzove akajairwa muhukuru hwezvivakwa.
Kuzvishandisa
Nei zvichikosha
- Kumhanya uye chiyero. Mutauro workflows inogona kufamba nekukurumidza pasina kupira kuenderana.
- Svika uye svika. Inopamhidzira kupinda mumitauro yese nemataera ekutaurirana.
- Sarudzo dzakajeka. Zvikwata zvinogona kupedza nguva yakawanda pakutonga uku otomatiki ichibata kudzokorora.
Chenjera
- Chokwadi chehuroyi chinogona kupinda chinyararire mishumo, kuyerera kwetsigiro, kana tsvakiridzo.
- Kunzwa nekukasira kunogona kugadzira mhedzisiro isingaenderane pane zvikumbiro zvakafanana.
- Sensitive text data inogona kuburitswa kana zvidhiraivho zvisina kusimba.
Matanho anotevera
- Tsanangura chimiro chekubuda, toni, uye mhando zviyero usati waburitsa.
- Mhinduro dzepasi neakavimbika masosi pese pazvine basa.
- Chengetedza ongororo yekuongorora yemunhu kune yakakwira-stake zvinobuda.
- Tevera maitiro ekutadza uye dzidzisazve kukurudzira kana mafambiro ebasa nguva nenguva.
Ramba Uchiongorora
Tsamba yemahara
Tora iyo yemazuva ese AI pfupiso
Nhatu dzakasimbiswa dzeAI nyaya mangwanani ega ega pakati pevhiki, dzakanyorwa neChirungu chakajeka. Yemahara zvachose, hapana ads.
Imwe email zuva rega rega revhiki. Bvisa kunyoresa kamwe chete. Hatimbotengese kana kugovera kero yako.
Zviedze iwe pachako
Take the Mixture of Depths quiz
Mhinduro yepakarepo pamhinduro yega yega, uye chitupa chinogovaniswa chine ID inobatika kana wapedza kosi.
Tsigira dzidzo yemahara yeAI. AI Understanding ndeye 501(c)(3) isingabatsiri - hapana ads, hapana paywall, zvachose. Ita chipo
Mibvunzo inowanzo bvunzwa
What is Mixture of Depths?
Musanganiswa Wezvakadzama (MoD) inoita kuti transformer ipedze huwandu hwakasiyana hwemakombuta pane akasiyana tokeni, ichiendesa chete 'akakosha' tokeni kuburikidza nechero layer inorema computation. Iyo inocheka mutengo wekugadzirisa zviri nyore tokens uchichengeta yakagadziriswa, inofanotaurwa compute bhajeti.
Chii chinonzi Musanganiswa Wezvakadzama unosiyana pane tokens?
MoD inofamba chete mamwe ma tokens kuburikidza nechero layer inorema computation, saka akasiyana tokens anowana kudzika kwakasiyana.
MoD inochengeta sei bhajeti rayo rekufungidzira?
Kusarudza yakagadziriswa yepamusoro-k yematokeni pa block caps FLOPs uye inochengeta tensor maumbirwo akasimba, inova inoshamwaridzana nehardware.
Chii chinoitika kune tokens iyo router haina sarudza pane yakapihwa block?
Matokeni asina kusarudzwa anopfuura komputa yeblock uye anotakurwa kumberi asina kuchinjika nenzira yasara.
Which lab published the Mixture of Depths technique?
Musanganiswa Wezvakadzama wakaunzwa ne Google DeepMind mubepa ra2024 pane dynamic transformer compute.
Chii chinopa kusanganisa MoD neMusanganiswa-we-Nyanzvi?
Kubatanidza kudzika kwenzira nehunyanzvi nzira inoburitsa 'MoDE' modhi iyo mamiriro anoteedzera pane ese maturu uye nyanzvi.
Ramba uchidzidza
Mazano anoenderana
Mamwe madhairekitori akasarudzirwa nyaya iyi