bitsandbytes/pr_1925/en/_app/immutable/nodes/8.344e170b.js

import{s as Q,n as B,o as D}from"../chunks/scheduler.8a2cc2fa.js";import{S as F,i as J,e as c,s,c as T,h as K,a as b,d as a,b as l,f as G,g as P,j as S,k as I,l as N,m as n,n as M,t as E,o as k,p as H}from"../chunks/index.7079e750.js";import{C as V,H as O,E as W}from"../chunks/MermaidChart.svelte_svelte_type_style_lang.0d032bd2.js";function X(A){let i,g,$,y,o,_,r,x,m,R="bitsandbytes enables accessible large language models via k-bit quantization for PyTorch. bitsandbytes provides three main features for dramatically reducing memory consumption for inference and training:",w,f,j="<li>8-bit optimizers uses block-wise quantization to maintain 32-bit performance at a small fraction of the memory cost.</li> <li>LLM.int8() or 8-bit quantization enables large language model inference with only half the required memory and without any performance degradation. This method is based on vector-wise quantization to quantize most features to 8-bits and separately treating outliers with 16-bit matrix multiplication.</li> <li>QLoRA or 4-bit quantization enables large language model training with several memory-saving techniques that don’t compromise performance. This method quantizes a model to 4-bits and inserts a small set of trainable low-rank adaptation (LoRA) weights to allow training.</li>",v,u,z,p,U="bitsandbytes is MIT licensed.",q,d,L,h,C;return o=new V({props:{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"}}),r=new O({props:{title:"bitsandbytes",local:"bitsandbytes",headingTag:"h1"}}),u=new O({props:{title:"License",local:"license",headingTag:"h1"}}),d=new W({props:{source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/main/docs/source/index.mdx"}}),{c(){i=c("meta"),g=s(),$=c("p"),y=s(),T(o.$$.fragment),_=s(),T(r.$$.fragment),x=s(),m=c("p"),m.textContent=R,w=s(),f=c("ul"),f.innerHTML=j,v=s(),T(u.$$.fragment),z=s(),p=c("p"),p.textContent=U,q=s(),T(d.$$.fragment),L=s(),h=c("p"),this.h()},l(t){const e=K("svelte-u9bgzb",document.head);i=b(e,"META",{name:!0,content:!0}),e.forEach(a),g=l(t),$=b(t,"P",{}),G($).forEach(a),y=l(t),P(o.$$.fragment,t),_=l(t),P(r.$$.fragment,t),x=l(t),m=b(t,"P",{"data-svelte-h":!0}),S(m)!=="svelte-18pq57t"&&(m.textContent=R),w=l(t),f=b(t,"UL",{"data-svelte-h":!0}),S(f)!=="svelte-2c8gj9"&&(f.innerHTML=j),v=l(t),P(u.$$.fragment,t),z=l(t),p=b(t,"P",{"data-svelte-h":!0}),S(p)!=="svelte-zzmy3r"&&(p.textContent=U),q=l(t),P(d.$$.fragment,t),L=l(t),h=b(t,"P",{}),G(h).forEach(a),this.h()},h(){I(i,"name","hf:doc:metadata"),I(i,"content",Y)},m(t,e){N(document.head,i),n(t,g,e),n(t,$,e),n(t,y,e),M(o,t,e),n(t,_,e),M(r,t,e),n(t,x,e),n(t,m,e),n(t,w,e),n(t,f,e),n(t,v,e),M(u,t,e),n(t,z,e),n(t,p,e),n(t,q,e),M(d,t,e),n(t,L,e),n(t,h,e),C=!0},p:B,i(t){C||(E(o.$$.fragment,t),E(r.$$.fragment,t),E(u.$$.fragment,t),E(d.$$.fragment,t),C=!0)},o(t){k(o.$$.fragment,t),k(r.$$.fragment,t),k(u.$$.fragment,t),k(d.$$.fragment,t),C=!1},d(t){t&&(a(g),a($),a(y),a(_),a(x),a(m),a(w),a(f),a(v),a(z),a(p),a(q),a(L),a(h)),a(i),H(o,t),H(r,t),H(u,t),H(d,t)}}}const Y='{"title":"bitsandbytes","local":"bitsandbytes","sections":[],"depth":1}';function Z(A){return D(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class nt extends F{constructor(i){super(),J(this,i,Z,X,Q,{})}}export{nt as component};