Sample text generated by RNN with different cell types, state size and layers.
Cell Type | State Size | Layers | Chars | Sample Text |
---|---|---|---|---|
Vanilla | 256 | 1 | 100 | n ungerking I dity ithatun. dik cy be n. YC. bikntan ing frditunthan dichit. buereg yer Hs ren Ifentre rmyo |
Vanilla | 256 | 1 | 1000 | I dereree ing I is e nd ee feng Des k th anewing ft kng I ikn rite dng Ing ky ine bey f ding rd 30: k. pe cie—t p inent be—thny ingingors ng (obentecing tinty peng uny but VCs If re dift.\n enrik ng ha it dine idinguat y ikenthecucu ci.\n Mat: ceringunt) pr"ewh at ple ding be ingntaingery 20% ferknguthe dor din bunt res res ere be peck perbewit y dindirrine r" be kin omuthan ing ututeng i be hend Wat bu bent fithafe 20% ng rery inknd I reriney nt Hy ingangh ding cin ithe ikinere 24]"ingend junthat ikit ines cit ang ut bute d. ngeng unt ngesfie ckerey ungrintiingo nguth catharingangner dimingunthingonghng tun I dingonorkid Weng antt intuaet. u Thakikikngrepent rikid Dakitut inghangesuatten in enve pl ngerenout y ithat at int tes buntutuntuth s dorifren cevidind ng? ng unllent utut pes peentha Yenge 30 d f ug tre chadele nd.\n M ikngut pere de be—t yod anga dinging s. cuntes ding besees Dthatyithent sere u nghatu igre inthat barey ppepingit ink fingatuthingh ory psere it pe cchat \n |
GRU | 256 | 1 | 100 | tair and think we'd stick in characters styling is usually \n presumably notice a \n despite early to work |
GRU | 256 | 1 | 1000 | eater timen actually\n true to discover it always tend.\n ([5]\n It's after for\n the general.[5]\n [22] Investors could try, want what's work for this lubb taxity to work on instanding their proof to catation caused as much.I vidents in the datts, I\n vary and debt to the quality of startups, and\n will hand give, that's while the technology called, it in\n your company,\n and that never asked them.One of the\n third enormouslined as a cause\n painting of starting a couldn't asked ourselves are much the mathem losed to talk foods, if analogiam decovered applications law in larger to take out\n people you pretend their point returned by point. And hard. When you get bookness. Most one\n seemed back—which look like what cook become language does and\n the\n things to recommend us to severieness.72, and some class of a thinking stalpped that food.It would everyone aren't. And years all Forg works of consequences. It matter in wealth when at facto is not desires attack a thirty of the eminent question is the still\n |
GRU | 1024 | 3 | 100 | *6] But the business \n I see, for excars. But more problems you should have a key \n valuations, but they |
GRU | 1024 | 3 | 1000 | Partiple\n , 20Don't current trouble since, but if it was competitors, \n the bottom,\n and papers; a can change your clife, the\n hypotheticians are meant things. [5]That's ins if it never as a third.Steames. This arounding resourcessful\n startups used by some\n working. It's not how Aprilliant members. Difference should probably say a sales are all talking a sense of investments\n which was\n probably always work; simply itself as backgrounded demand,"What show Applace too fright? \n runct thing that suits ride and it is active to getcriptible a media lower of third.Unvestors will be yolarh turning your curvs. Obcortinaril. Apple explanaties ereal currounding dangerously\n difficult as an argument no more people is million outside.Or startup, we've\n wrong bottlects tell when you'll be publish\n men we wrote the founders yet it were. And yet what's the valuable of it was about my madents. But we even though it.\n But if it is "quickly, you'll have to be simply\n happening ones,\n you already best startup.\n |
LSTM | 512 | 1 | 100 | hoodsly'd buged by mean aren't to strange is in geveralong is \n they never an own is, not just thes gra |
LSTM | 512 | 1 | 1000 | s merely and seem to lues that to\n started is\n about whether your experts of the way most\n of rungion software you want to be user as a nowical Mimportant That's always of ongooy\n plagering,.Whereeas at access in 17% to try, there trassly frompainto ration correct 1006\n of reading, a startupt cander to preventing selvessayee.L. Munfrilion with a lit before there are propabally tooks, betaxter is dispeciend to may\n and mank.For least harded that\n Intlined and any knows far are invissauly\n be a position of infures—Vark\n question and its strigghtweld, "R]Y9ut the wrote know good take openatalary Asmers been running search eef they see had two -Comperse out\n of probabaries\n as idea to expected the suffore serveship exppe, in increases where\n werentely to change offered in your imatives against else common you counds for thesel,\n and soorsistril is not hist, and in the work for what you don't navers, get fire is a supports is decides are not to make a write are lose. Art that part of insixply to get loo\n |
LSTM | 512 | 2 | 100 | Guter. If a valuation was to learn to start one. Writers, here is photost Someone who was later as |
LSTM | 512 | 2 | 1000 | Our Abinlinution is tend's so, it's much last to zero familization in the high school society Microsoft we could be meetings change of hemorable to start a startup\n that take \n because server-startups machine whatever they have getting economics end-gradea is decisions like the religion a high-methlity, C8N Disps?And the founders who seem no was the\n gain of beil. Or not replacing. It's not stupide than help-baround togethers. The related\n wealth.Be aid it, not is tried along\n job what's what a startup hubs.The web liked an impressively big\n people the way to be some downbar office themselves gradually be: it's ton, you disappoint it seffect the organization\n being the last time when we were always be founded, eithin the last in creating business are raising geats in an adbitious problem is friex, he may make a company have to started out to bringing himself in big treen of users.\n Sundrad mericonds who've never olders were, people want. [15]]4. Since the reason\n\n school. |