Buckets:

hf-doc-build/doc / optimum-neuron /main /en /_app /pages /package_reference /export.mdx-hf-doc-builder.js
rtrm's picture
download
raw
22.7 kB
import{S as zr,i as Wr,s as Jr,e as a,k as s,w as ke,t as l,M as jr,c as i,d as e,m as c,a as r,x as ge,h as n,b as p,G as t,g as u,y as ye,q as De,o as Re,B as $e,v as Kr}from"../../chunks/vendor-hf-doc-builder.js";import{T as Qr}from"../../chunks/Tip-hf-doc-builder.js";import{I as vo}from"../../chunks/IconCopyLink-hf-doc-builder.js";function Vr(be){let m,N,v,_,D;return{c(){m=a("p"),N=l("More details for checking supported tasks "),v=a("a"),_=l("here"),D=l("."),this.h()},l(E){m=i(E,"P",{});var k=r(m);N=n(k,"More details for checking supported tasks "),v=i(k,"A",{href:!0,rel:!0});var q=r(v);_=n(q,"here"),q.forEach(e),D=n(k,"."),k.forEach(e),this.h()},h(){p(v,"href","https://huggingface.co/docs/optimum-neuron/guides/export_model#selecting-a-task"),p(v,"rel","nofollow")},m(E,k){u(E,m,k),t(m,N),t(m,v),t(v,_),t(m,D)},d(E){E&&e(m)}}}function Zr(be){let m,N,v,_,D,E,k,q,Eo,Ne,g,_o,F,xo,To,H,wo,ko,qe,R,B,Et,X,go,_t,yo,Be,x,Do,xt,Ro,$o,Tt,bo,No,wt,qo,Bo,Pe,P,Po,kt,Ao,Co,Ae,$,A,gt,G,Io,yt,So,Ce,ht,Lo,Ie,T,Dt,Mo,Oo,Rt,Fo,Ho,$t,Xo,Go,bt,Uo,Se,w,Yo,Nt,zo,Wo,qt,Jo,jo,Bt,Ko,Qo,Le,ut,Vo,Me,y,Pt,Zo,ta,At,ea,oa,U,aa,Ct,ia,ra,Oe,b,C,It,Y,la,St,na,Fe,I,Lt,z,Mt,sa,ca,Ot,fa,da,f,W,Ft,ha,ua,Ht,pa,ma,J,Xt,va,Ea,Gt,_a,xa,j,Ut,Ta,wa,Yt,ka,ga,K,zt,ya,Da,Wt,Ra,$a,Q,Jt,ba,Na,jt,qa,Ba,V,Kt,Pa,Aa,Qt,Ca,Ia,Z,Vt,Sa,La,Zt,Ma,Oa,tt,te,Fa,Ha,ee,Xa,Ga,et,oe,Ua,Ya,ae,za,Wa,ot,ie,Ja,ja,re,Ka,Qa,at,le,Va,Za,ne,ti,ei,it,se,oi,ai,ce,ii,ri,rt,fe,li,ni,de,si,ci,lt,he,fi,di,ue,hi,ui,nt,pe,pi,mi,me,vi,Ei,st,ve,_i,xi,Ee,Ti,wi,ct,_e,ki,gi,xe,yi,Di,ft,Te,Ri,$i,we,bi,He,S,Xe,pt,Ni,Ge;return E=new vo({}),X=new vo({}),G=new vo({}),Y=new vo({}),S=new Qr({props:{$$slots:{default:[Vr]},$$scope:{ctx:be}}}),{c(){m=a("meta"),N=s(),v=a("h1"),_=a("a"),D=a("span"),ke(E.$$.fragment),k=s(),q=a("span"),Eo=l("Inferentia Exporter"),Ne=s(),g=a("p"),_o=l("You can export a PyTorch model to Neuron with \u{1F917} Optimum to run inference on AWS "),F=a("a"),xo=l("Inferntia 1"),To=l(`
and `),H=a("a"),wo=l("Inferentia 2"),ko=l("."),qe=s(),R=a("h2"),B=a("a"),Et=a("span"),ke(X.$$.fragment),go=s(),_t=a("span"),yo=l("Export functions"),Be=s(),x=a("p"),Do=l("There is an export function for each generation of the Inferentia accelerator, "),xt=a("code"),Ro=l("export_neuron"),$o=l(`
for INF1 and `),Tt=a("code"),bo=l("export_neuronx"),No=l(" on INF2, but you will be able to use directly the export function "),wt=a("code"),qo=l("export"),Bo=l(`, which will select the proper
exporting function according to the environment.`),Pe=s(),P=a("p"),Po=l("Besides, you can check if the exported model is valid via "),kt=a("code"),Ao=l("validate_model_outputs"),Co=l(`, which compares
the compiled model\u2019s output on Neuron devices to the PyTorch model\u2019s output on CPU.`),Ae=s(),$=a("h2"),A=a("a"),gt=a("span"),ke(G.$$.fragment),Io=s(),yt=a("span"),So=l("Configuration classes for Neuron exports"),Ce=s(),ht=a("p"),Lo=l("Exporting a PyTorch model to a Neuron compiled model involves specifying:"),Ie=s(),T=a("ol"),Dt=a("li"),Mo=l("The input names."),Oo=s(),Rt=a("li"),Fo=l("The output names."),Ho=s(),$t=a("li"),Xo=l("The dummy inputs used to trace the model. This is needed by the Neuron Compiler to record the computational graph and convert it to a TorchScript module."),Go=s(),bt=a("li"),Uo=l("The compilation arguments used to control the trade-off between hardware efficiency (latency, throughput) and accuracy."),Se=s(),w=a("p"),Yo=l("Depending on the choice of model and task, we represent the data above with "),Nt=a("em"),zo=l("configuration classes"),Wo=l(`. Each configuration class is associated with
a specific model architecture, and follows the naming convention `),qt=a("code"),Jo=l("ArchitectureNameNeuronConfig"),jo=l(`. For instance, the configuration which specifies the Neuron
export of BERT models is `),Bt=a("code"),Ko=l("BertNeuronConfig"),Qo=l("."),Le=s(),ut=a("p"),Vo=l("Since many architectures share similar properties for their Neuron configuration, \u{1F917} Optimum adopts a 3-level class hierarchy:"),Me=s(),y=a("ol"),Pt=a("li"),Zo=l("Abstract and generic base classes. These handle all the fundamental features, while being agnostic to the modality (text, image, audio, etc)."),ta=s(),At=a("li"),ea=l(`Middle-end classes. These are aware of the modality, but multiple can exist for the same modality depending on the inputs they support.
They specify which input generators should be used for the dummy inputs, but remain model-agnostic.`),oa=s(),U=a("li"),aa=l("Model-specific classes like the "),Ct=a("code"),ia=l("BertNeuronConfig"),ra=l(" mentioned above. These are the ones actually used to export models."),Oe=s(),b=a("h2"),C=a("a"),It=a("span"),ke(Y.$$.fragment),la=s(),St=a("span"),na=l("Supported architectures"),Fe=s(),I=a("table"),Lt=a("thead"),z=a("tr"),Mt=a("th"),sa=l("Architecture"),ca=s(),Ot=a("th"),fa=l("Task"),da=s(),f=a("tbody"),W=a("tr"),Ft=a("td"),ha=l("ALBERT"),ua=s(),Ht=a("td"),pa=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),ma=s(),J=a("tr"),Xt=a("td"),va=l("BERT"),Ea=s(),Gt=a("td"),_a=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),xa=s(),j=a("tr"),Ut=a("td"),Ta=l("CamemBERT"),wa=s(),Yt=a("td"),ka=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),ga=s(),K=a("tr"),zt=a("td"),ya=l("ConvBERT"),Da=s(),Wt=a("td"),Ra=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),$a=s(),Q=a("tr"),Jt=a("td"),ba=l("DeBERTa (INF2 only)"),Na=s(),jt=a("td"),qa=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Ba=s(),V=a("tr"),Kt=a("td"),Pa=l("DeBERTa-v2 (INF2 only)"),Aa=s(),Qt=a("td"),Ca=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Ia=s(),Z=a("tr"),Vt=a("td"),Sa=l("DistilBERT"),La=s(),Zt=a("td"),Ma=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Oa=s(),tt=a("tr"),te=a("td"),Fa=l("ELECTRA"),Ha=s(),ee=a("td"),Xa=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Ga=s(),et=a("tr"),oe=a("td"),Ua=l("FlauBERT"),Ya=s(),ae=a("td"),za=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Wa=s(),ot=a("tr"),ie=a("td"),Ja=l("GPT2"),ja=s(),re=a("td"),Ka=l("text-generation"),Qa=s(),at=a("tr"),le=a("td"),Va=l("MobileBERT"),Za=s(),ne=a("td"),ti=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),ei=s(),it=a("tr"),se=a("td"),oi=l("MPNet"),ai=s(),ce=a("td"),ii=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),ri=s(),rt=a("tr"),fe=a("td"),li=l("RoBERTa"),ni=s(),de=a("td"),si=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),ci=s(),lt=a("tr"),he=a("td"),fi=l("RoFormer"),di=s(),ue=a("td"),hi=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),ui=s(),nt=a("tr"),pe=a("td"),pi=l("XLM"),mi=s(),me=a("td"),vi=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Ei=s(),st=a("tr"),ve=a("td"),_i=l("XLM-RoBERTa"),xi=s(),Ee=a("td"),Ti=l("feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),wi=s(),ct=a("tr"),_e=a("td"),ki=l("Stable Diffusion"),gi=s(),xe=a("td"),yi=l("text-to-image, image-to-image, inpaint"),Di=s(),ft=a("tr"),Te=a("td"),Ri=l("Stable Diffusion XL"),$i=s(),we=a("td"),bi=l("text-to-image"),He=s(),ke(S.$$.fragment),Xe=s(),pt=a("p"),Ni=l("More architectures coming soon, stay tuned! \u{1F680}"),this.h()},l(o){const d=jr('[data-svelte="svelte-1phssyn"]',document.head);m=i(d,"META",{name:!0,content:!0}),d.forEach(e),N=c(o),v=i(o,"H1",{class:!0});var dt=r(v);_=i(dt,"A",{id:!0,class:!0,href:!0});var qi=r(_);D=i(qi,"SPAN",{});var Bi=r(D);ge(E.$$.fragment,Bi),Bi.forEach(e),qi.forEach(e),k=c(dt),q=i(dt,"SPAN",{});var Pi=r(q);Eo=n(Pi,"Inferentia Exporter"),Pi.forEach(e),dt.forEach(e),Ne=c(o),g=i(o,"P",{});var mt=r(g);_o=n(mt,"You can export a PyTorch model to Neuron with \u{1F917} Optimum to run inference on AWS "),F=i(mt,"A",{href:!0,rel:!0});var Ai=r(F);xo=n(Ai,"Inferntia 1"),Ai.forEach(e),To=n(mt,`
and `),H=i(mt,"A",{href:!0,rel:!0});var Ci=r(H);wo=n(Ci,"Inferentia 2"),Ci.forEach(e),ko=n(mt,"."),mt.forEach(e),qe=c(o),R=i(o,"H2",{class:!0});var Ue=r(R);B=i(Ue,"A",{id:!0,class:!0,href:!0});var Ii=r(B);Et=i(Ii,"SPAN",{});var Si=r(Et);ge(X.$$.fragment,Si),Si.forEach(e),Ii.forEach(e),go=c(Ue),_t=i(Ue,"SPAN",{});var Li=r(_t);yo=n(Li,"Export functions"),Li.forEach(e),Ue.forEach(e),Be=c(o),x=i(o,"P",{});var L=r(x);Do=n(L,"There is an export function for each generation of the Inferentia accelerator, "),xt=i(L,"CODE",{});var Mi=r(xt);Ro=n(Mi,"export_neuron"),Mi.forEach(e),$o=n(L,`
for INF1 and `),Tt=i(L,"CODE",{});var Oi=r(Tt);bo=n(Oi,"export_neuronx"),Oi.forEach(e),No=n(L," on INF2, but you will be able to use directly the export function "),wt=i(L,"CODE",{});var Fi=r(wt);qo=n(Fi,"export"),Fi.forEach(e),Bo=n(L,`, which will select the proper
exporting function according to the environment.`),L.forEach(e),Pe=c(o),P=i(o,"P",{});var Ye=r(P);Po=n(Ye,"Besides, you can check if the exported model is valid via "),kt=i(Ye,"CODE",{});var Hi=r(kt);Ao=n(Hi,"validate_model_outputs"),Hi.forEach(e),Co=n(Ye,`, which compares
the compiled model\u2019s output on Neuron devices to the PyTorch model\u2019s output on CPU.`),Ye.forEach(e),Ae=c(o),$=i(o,"H2",{class:!0});var ze=r($);A=i(ze,"A",{id:!0,class:!0,href:!0});var Xi=r(A);gt=i(Xi,"SPAN",{});var Gi=r(gt);ge(G.$$.fragment,Gi),Gi.forEach(e),Xi.forEach(e),Io=c(ze),yt=i(ze,"SPAN",{});var Ui=r(yt);So=n(Ui,"Configuration classes for Neuron exports"),Ui.forEach(e),ze.forEach(e),Ce=c(o),ht=i(o,"P",{});var Yi=r(ht);Lo=n(Yi,"Exporting a PyTorch model to a Neuron compiled model involves specifying:"),Yi.forEach(e),Ie=c(o),T=i(o,"OL",{});var M=r(T);Dt=i(M,"LI",{});var zi=r(Dt);Mo=n(zi,"The input names."),zi.forEach(e),Oo=c(M),Rt=i(M,"LI",{});var Wi=r(Rt);Fo=n(Wi,"The output names."),Wi.forEach(e),Ho=c(M),$t=i(M,"LI",{});var Ji=r($t);Xo=n(Ji,"The dummy inputs used to trace the model. This is needed by the Neuron Compiler to record the computational graph and convert it to a TorchScript module."),Ji.forEach(e),Go=c(M),bt=i(M,"LI",{});var ji=r(bt);Uo=n(ji,"The compilation arguments used to control the trade-off between hardware efficiency (latency, throughput) and accuracy."),ji.forEach(e),M.forEach(e),Se=c(o),w=i(o,"P",{});var O=r(w);Yo=n(O,"Depending on the choice of model and task, we represent the data above with "),Nt=i(O,"EM",{});var Ki=r(Nt);zo=n(Ki,"configuration classes"),Ki.forEach(e),Wo=n(O,`. Each configuration class is associated with
a specific model architecture, and follows the naming convention `),qt=i(O,"CODE",{});var Qi=r(qt);Jo=n(Qi,"ArchitectureNameNeuronConfig"),Qi.forEach(e),jo=n(O,`. For instance, the configuration which specifies the Neuron
export of BERT models is `),Bt=i(O,"CODE",{});var Vi=r(Bt);Ko=n(Vi,"BertNeuronConfig"),Vi.forEach(e),Qo=n(O,"."),O.forEach(e),Le=c(o),ut=i(o,"P",{});var Zi=r(ut);Vo=n(Zi,"Since many architectures share similar properties for their Neuron configuration, \u{1F917} Optimum adopts a 3-level class hierarchy:"),Zi.forEach(e),Me=c(o),y=i(o,"OL",{});var vt=r(y);Pt=i(vt,"LI",{});var tr=r(Pt);Zo=n(tr,"Abstract and generic base classes. These handle all the fundamental features, while being agnostic to the modality (text, image, audio, etc)."),tr.forEach(e),ta=c(vt),At=i(vt,"LI",{});var er=r(At);ea=n(er,`Middle-end classes. These are aware of the modality, but multiple can exist for the same modality depending on the inputs they support.
They specify which input generators should be used for the dummy inputs, but remain model-agnostic.`),er.forEach(e),oa=c(vt),U=i(vt,"LI",{});var We=r(U);aa=n(We,"Model-specific classes like the "),Ct=i(We,"CODE",{});var or=r(Ct);ia=n(or,"BertNeuronConfig"),or.forEach(e),ra=n(We," mentioned above. These are the ones actually used to export models."),We.forEach(e),vt.forEach(e),Oe=c(o),b=i(o,"H2",{class:!0});var Je=r(b);C=i(Je,"A",{id:!0,class:!0,href:!0});var ar=r(C);It=i(ar,"SPAN",{});var ir=r(It);ge(Y.$$.fragment,ir),ir.forEach(e),ar.forEach(e),la=c(Je),St=i(Je,"SPAN",{});var rr=r(St);na=n(rr,"Supported architectures"),rr.forEach(e),Je.forEach(e),Fe=c(o),I=i(o,"TABLE",{});var je=r(I);Lt=i(je,"THEAD",{});var lr=r(Lt);z=i(lr,"TR",{});var Ke=r(z);Mt=i(Ke,"TH",{});var nr=r(Mt);sa=n(nr,"Architecture"),nr.forEach(e),ca=c(Ke),Ot=i(Ke,"TH",{});var sr=r(Ot);fa=n(sr,"Task"),sr.forEach(e),Ke.forEach(e),lr.forEach(e),da=c(je),f=i(je,"TBODY",{});var h=r(f);W=i(h,"TR",{});var Qe=r(W);Ft=i(Qe,"TD",{});var cr=r(Ft);ha=n(cr,"ALBERT"),cr.forEach(e),ua=c(Qe),Ht=i(Qe,"TD",{});var fr=r(Ht);pa=n(fr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),fr.forEach(e),Qe.forEach(e),ma=c(h),J=i(h,"TR",{});var Ve=r(J);Xt=i(Ve,"TD",{});var dr=r(Xt);va=n(dr,"BERT"),dr.forEach(e),Ea=c(Ve),Gt=i(Ve,"TD",{});var hr=r(Gt);_a=n(hr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),hr.forEach(e),Ve.forEach(e),xa=c(h),j=i(h,"TR",{});var Ze=r(j);Ut=i(Ze,"TD",{});var ur=r(Ut);Ta=n(ur,"CamemBERT"),ur.forEach(e),wa=c(Ze),Yt=i(Ze,"TD",{});var pr=r(Yt);ka=n(pr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),pr.forEach(e),Ze.forEach(e),ga=c(h),K=i(h,"TR",{});var to=r(K);zt=i(to,"TD",{});var mr=r(zt);ya=n(mr,"ConvBERT"),mr.forEach(e),Da=c(to),Wt=i(to,"TD",{});var vr=r(Wt);Ra=n(vr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),vr.forEach(e),to.forEach(e),$a=c(h),Q=i(h,"TR",{});var eo=r(Q);Jt=i(eo,"TD",{});var Er=r(Jt);ba=n(Er,"DeBERTa (INF2 only)"),Er.forEach(e),Na=c(eo),jt=i(eo,"TD",{});var _r=r(jt);qa=n(_r,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),_r.forEach(e),eo.forEach(e),Ba=c(h),V=i(h,"TR",{});var oo=r(V);Kt=i(oo,"TD",{});var xr=r(Kt);Pa=n(xr,"DeBERTa-v2 (INF2 only)"),xr.forEach(e),Aa=c(oo),Qt=i(oo,"TD",{});var Tr=r(Qt);Ca=n(Tr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Tr.forEach(e),oo.forEach(e),Ia=c(h),Z=i(h,"TR",{});var ao=r(Z);Vt=i(ao,"TD",{});var wr=r(Vt);Sa=n(wr,"DistilBERT"),wr.forEach(e),La=c(ao),Zt=i(ao,"TD",{});var kr=r(Zt);Ma=n(kr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),kr.forEach(e),ao.forEach(e),Oa=c(h),tt=i(h,"TR",{});var io=r(tt);te=i(io,"TD",{});var gr=r(te);Fa=n(gr,"ELECTRA"),gr.forEach(e),Ha=c(io),ee=i(io,"TD",{});var yr=r(ee);Xa=n(yr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),yr.forEach(e),io.forEach(e),Ga=c(h),et=i(h,"TR",{});var ro=r(et);oe=i(ro,"TD",{});var Dr=r(oe);Ua=n(Dr,"FlauBERT"),Dr.forEach(e),Ya=c(ro),ae=i(ro,"TD",{});var Rr=r(ae);za=n(Rr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Rr.forEach(e),ro.forEach(e),Wa=c(h),ot=i(h,"TR",{});var lo=r(ot);ie=i(lo,"TD",{});var $r=r(ie);Ja=n($r,"GPT2"),$r.forEach(e),ja=c(lo),re=i(lo,"TD",{});var br=r(re);Ka=n(br,"text-generation"),br.forEach(e),lo.forEach(e),Qa=c(h),at=i(h,"TR",{});var no=r(at);le=i(no,"TD",{});var Nr=r(le);Va=n(Nr,"MobileBERT"),Nr.forEach(e),Za=c(no),ne=i(no,"TD",{});var qr=r(ne);ti=n(qr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),qr.forEach(e),no.forEach(e),ei=c(h),it=i(h,"TR",{});var so=r(it);se=i(so,"TD",{});var Br=r(se);oi=n(Br,"MPNet"),Br.forEach(e),ai=c(so),ce=i(so,"TD",{});var Pr=r(ce);ii=n(Pr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Pr.forEach(e),so.forEach(e),ri=c(h),rt=i(h,"TR",{});var co=r(rt);fe=i(co,"TD",{});var Ar=r(fe);li=n(Ar,"RoBERTa"),Ar.forEach(e),ni=c(co),de=i(co,"TD",{});var Cr=r(de);si=n(Cr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Cr.forEach(e),co.forEach(e),ci=c(h),lt=i(h,"TR",{});var fo=r(lt);he=i(fo,"TD",{});var Ir=r(he);fi=n(Ir,"RoFormer"),Ir.forEach(e),di=c(fo),ue=i(fo,"TD",{});var Sr=r(ue);hi=n(Sr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Sr.forEach(e),fo.forEach(e),ui=c(h),nt=i(h,"TR",{});var ho=r(nt);pe=i(ho,"TD",{});var Lr=r(pe);pi=n(Lr,"XLM"),Lr.forEach(e),mi=c(ho),me=i(ho,"TD",{});var Mr=r(me);vi=n(Mr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Mr.forEach(e),ho.forEach(e),Ei=c(h),st=i(h,"TR",{});var uo=r(st);ve=i(uo,"TD",{});var Or=r(ve);_i=n(Or,"XLM-RoBERTa"),Or.forEach(e),xi=c(uo),Ee=i(uo,"TD",{});var Fr=r(Ee);Ti=n(Fr,"feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification"),Fr.forEach(e),uo.forEach(e),wi=c(h),ct=i(h,"TR",{});var po=r(ct);_e=i(po,"TD",{});var Hr=r(_e);ki=n(Hr,"Stable Diffusion"),Hr.forEach(e),gi=c(po),xe=i(po,"TD",{});var Xr=r(xe);yi=n(Xr,"text-to-image, image-to-image, inpaint"),Xr.forEach(e),po.forEach(e),Di=c(h),ft=i(h,"TR",{});var mo=r(ft);Te=i(mo,"TD",{});var Gr=r(Te);Ri=n(Gr,"Stable Diffusion XL"),Gr.forEach(e),$i=c(mo),we=i(mo,"TD",{});var Ur=r(we);bi=n(Ur,"text-to-image"),Ur.forEach(e),mo.forEach(e),h.forEach(e),je.forEach(e),He=c(o),ge(S.$$.fragment,o),Xe=c(o),pt=i(o,"P",{});var Yr=r(pt);Ni=n(Yr,"More architectures coming soon, stay tuned! \u{1F680}"),Yr.forEach(e),this.h()},h(){p(m,"name","hf:doc:metadata"),p(m,"content",JSON.stringify(tl)),p(_,"id","inferentia-exporter"),p(_,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),p(_,"href","#inferentia-exporter"),p(v,"class","relative group"),p(F,"href","https://aws.amazon.com/ec2/instance-types/inf1/"),p(F,"rel","nofollow"),p(H,"href","https://aws.amazon.com/ec2/instance-types/inf2/"),p(H,"rel","nofollow"),p(B,"id","export-functions"),p(B,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),p(B,"href","#export-functions"),p(R,"class","relative group"),p(A,"id","configuration-classes-for-neuron-exports"),p(A,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),p(A,"href","#configuration-classes-for-neuron-exports"),p($,"class","relative group"),p(C,"id","supported-architectures"),p(C,"class","header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full"),p(C,"href","#supported-architectures"),p(b,"class","relative group")},m(o,d){t(document.head,m),u(o,N,d),u(o,v,d),t(v,_),t(_,D),ye(E,D,null),t(v,k),t(v,q),t(q,Eo),u(o,Ne,d),u(o,g,d),t(g,_o),t(g,F),t(F,xo),t(g,To),t(g,H),t(H,wo),t(g,ko),u(o,qe,d),u(o,R,d),t(R,B),t(B,Et),ye(X,Et,null),t(R,go),t(R,_t),t(_t,yo),u(o,Be,d),u(o,x,d),t(x,Do),t(x,xt),t(xt,Ro),t(x,$o),t(x,Tt),t(Tt,bo),t(x,No),t(x,wt),t(wt,qo),t(x,Bo),u(o,Pe,d),u(o,P,d),t(P,Po),t(P,kt),t(kt,Ao),t(P,Co),u(o,Ae,d),u(o,$,d),t($,A),t(A,gt),ye(G,gt,null),t($,Io),t($,yt),t(yt,So),u(o,Ce,d),u(o,ht,d),t(ht,Lo),u(o,Ie,d),u(o,T,d),t(T,Dt),t(Dt,Mo),t(T,Oo),t(T,Rt),t(Rt,Fo),t(T,Ho),t(T,$t),t($t,Xo),t(T,Go),t(T,bt),t(bt,Uo),u(o,Se,d),u(o,w,d),t(w,Yo),t(w,Nt),t(Nt,zo),t(w,Wo),t(w,qt),t(qt,Jo),t(w,jo),t(w,Bt),t(Bt,Ko),t(w,Qo),u(o,Le,d),u(o,ut,d),t(ut,Vo),u(o,Me,d),u(o,y,d),t(y,Pt),t(Pt,Zo),t(y,ta),t(y,At),t(At,ea),t(y,oa),t(y,U),t(U,aa),t(U,Ct),t(Ct,ia),t(U,ra),u(o,Oe,d),u(o,b,d),t(b,C),t(C,It),ye(Y,It,null),t(b,la),t(b,St),t(St,na),u(o,Fe,d),u(o,I,d),t(I,Lt),t(Lt,z),t(z,Mt),t(Mt,sa),t(z,ca),t(z,Ot),t(Ot,fa),t(I,da),t(I,f),t(f,W),t(W,Ft),t(Ft,ha),t(W,ua),t(W,Ht),t(Ht,pa),t(f,ma),t(f,J),t(J,Xt),t(Xt,va),t(J,Ea),t(J,Gt),t(Gt,_a),t(f,xa),t(f,j),t(j,Ut),t(Ut,Ta),t(j,wa),t(j,Yt),t(Yt,ka),t(f,ga),t(f,K),t(K,zt),t(zt,ya),t(K,Da),t(K,Wt),t(Wt,Ra),t(f,$a),t(f,Q),t(Q,Jt),t(Jt,ba),t(Q,Na),t(Q,jt),t(jt,qa),t(f,Ba),t(f,V),t(V,Kt),t(Kt,Pa),t(V,Aa),t(V,Qt),t(Qt,Ca),t(f,Ia),t(f,Z),t(Z,Vt),t(Vt,Sa),t(Z,La),t(Z,Zt),t(Zt,Ma),t(f,Oa),t(f,tt),t(tt,te),t(te,Fa),t(tt,Ha),t(tt,ee),t(ee,Xa),t(f,Ga),t(f,et),t(et,oe),t(oe,Ua),t(et,Ya),t(et,ae),t(ae,za),t(f,Wa),t(f,ot),t(ot,ie),t(ie,Ja),t(ot,ja),t(ot,re),t(re,Ka),t(f,Qa),t(f,at),t(at,le),t(le,Va),t(at,Za),t(at,ne),t(ne,ti),t(f,ei),t(f,it),t(it,se),t(se,oi),t(it,ai),t(it,ce),t(ce,ii),t(f,ri),t(f,rt),t(rt,fe),t(fe,li),t(rt,ni),t(rt,de),t(de,si),t(f,ci),t(f,lt),t(lt,he),t(he,fi),t(lt,di),t(lt,ue),t(ue,hi),t(f,ui),t(f,nt),t(nt,pe),t(pe,pi),t(nt,mi),t(nt,me),t(me,vi),t(f,Ei),t(f,st),t(st,ve),t(ve,_i),t(st,xi),t(st,Ee),t(Ee,Ti),t(f,wi),t(f,ct),t(ct,_e),t(_e,ki),t(ct,gi),t(ct,xe),t(xe,yi),t(f,Di),t(f,ft),t(ft,Te),t(Te,Ri),t(ft,$i),t(ft,we),t(we,bi),u(o,He,d),ye(S,o,d),u(o,Xe,d),u(o,pt,d),t(pt,Ni),Ge=!0},p(o,[d]){const dt={};d&2&&(dt.$$scope={dirty:d,ctx:o}),S.$set(dt)},i(o){Ge||(De(E.$$.fragment,o),De(X.$$.fragment,o),De(G.$$.fragment,o),De(Y.$$.fragment,o),De(S.$$.fragment,o),Ge=!0)},o(o){Re(E.$$.fragment,o),Re(X.$$.fragment,o),Re(G.$$.fragment,o),Re(Y.$$.fragment,o),Re(S.$$.fragment,o),Ge=!1},d(o){e(m),o&&e(N),o&&e(v),$e(E),o&&e(Ne),o&&e(g),o&&e(qe),o&&e(R),$e(X),o&&e(Be),o&&e(x),o&&e(Pe),o&&e(P),o&&e(Ae),o&&e($),$e(G),o&&e(Ce),o&&e(ht),o&&e(Ie),o&&e(T),o&&e(Se),o&&e(w),o&&e(Le),o&&e(ut),o&&e(Me),o&&e(y),o&&e(Oe),o&&e(b),$e(Y),o&&e(Fe),o&&e(I),o&&e(He),$e(S,o),o&&e(Xe),o&&e(pt)}}}const tl={local:"inferentia-exporter",sections:[{local:"export-functions",title:"Export functions"},{local:"configuration-classes-for-neuron-exports",title:"Configuration classes for Neuron exports"},{local:"supported-architectures",title:"Supported architectures"}],title:"Inferentia Exporter"};function el(be){return Kr(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class rl extends zr{constructor(m){super();Wr(this,m,el,Zr,Jr,{})}}export{rl as default,tl as metadata};

Xet Storage Details

Size:
22.7 kB
·
Xet hash:
fdc27f1f68cbb187a78c137778563d2133a7f3c35df1a37069da5cd7f107a394

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.