bitsandbytes/pr_1908/en/_app/immutable/nodes/19.843b17a2.js

import{s as me,n as be,o as pe}from"../chunks/scheduler.8a2cc2fa.js";import{S as le,i as _e,e as r,s as i,c as m,h as ce,a as d,d as e,b as n,f as g,g as b,j as D,k as h,l as o,m as s,n as p,t as l,o as _,p as c}from"../chunks/index.7079e750.js";import{C as ge,H as et,E as he}from"../chunks/MermaidChart.svelte_svelte_type_style_lang.2b7ce466.js";import{D as f}from"../chunks/Docstring.8c9a5003.js";function fe(te){let u,bt,dt,pt,N,lt,E,_t,L,ee='<a href="https://hf.co/papers/1711.05101" rel="nofollow">AdamW</a> is a variant of the <code>Adam</code> optimizer that separates weight decay from the gradient update based on the observation that the weight decay formulation is different when applied to <code>SGD</code> and <code>Adam</code>.',ct,j,ae="bitsandbytes also supports paged optimizers which take advantage of CUDAs unified memory to transfer memory from the GPU to the CPU when GPU memory is exhausted.",gt,I,ht,v,M,Ut,w,V,Ht,at,ie="Base AdamW optimizer.",ft,S,ut,y,G,kt,z,q,Bt,it,ne="8-bit AdamW optimizer.",vt,U,yt,$,H,Ot,P,k,Rt,nt,se="32-bit AdamW optimizer.",$t,B,At,A,O,Jt,T,R,Kt,st,oe="Paged AdamW optimizer.",Wt,J,xt,W,K,Qt,F,Q,Xt,ot,re="Paged 8-bit AdamW optimizer.",wt,X,zt,x,Y,Yt,C,Z,Zt,rt,de="Paged 32-bit AdamW optimizer.",Pt,tt,Tt,mt,Ft;return N=new ge({props:{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"}}),E=new et({props:{title:"AdamW",local:"adamw",headingTag:"h1"}}),I=new et({props:{title:"AdamW",local:"api-class ][ bitsandbytes.optim.AdamW",headingTag:"h2"}}),M=new f({props:{name:"class bitsandbytes.optim.AdamW",anchor:"bitsandbytes.optim.AdamW",parameters:[{name:"params",val:""},{name:"lr",val:" = 0.001"},{name:"betas",val:" = (0.9, 0.999)"},{name:"eps",val:" = 1e-08"},{name:"weight_decay",val:" = 0.01"},{name:"amsgrad",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"},{name:"is_paged",val:" = False"}],source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_1908/bitsandbytes/optim/adamw.py#L9"}}),V=new f({props:{name:"__init__",anchor:"bitsandbytes.optim.AdamW.__init__",parameters:[{name:"params",val:""},{name:"lr",val:" = 0.001"},{name:"betas",val:" = (0.9, 0.999)"},{name:"eps",val:" = 1e-08"},{name:"weight_decay",val:" = 0.01"},{name:"amsgrad",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"},{name:"is_paged",val:" = False"}],parametersDescription:[{anchor:"bitsandbytes.optim.AdamW.__init__.params",description:`<strong>params</strong> (<code>torch.Tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.AdamW.__init__.lr",description:`<strong>lr</strong> (<code>float</code>, defaults to 1e-3) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.AdamW.__init__.betas",description:`<strong>betas</strong> (<code>tuple(float, float)</code>, defaults to (0.9, 0.999)) &#x2014;
The beta values are the decay rates of the first and second-order moment of the optimizer.`,name:"betas"},{anchor:"bitsandbytes.optim.AdamW.__init__.eps",description:`<strong>eps</strong> (<code>float</code>, defaults to 1e-8) &#x2014;
The epsilon value prevents division by zero in the optimizer.`,name:"eps"},{anchor:"bitsandbytes.optim.AdamW.__init__.weight_decay",description:`<strong>weight_decay</strong> (<code>float</code>, defaults to 1e-2) &#x2014;
The weight decay value for the optimizer.`,name:"weight_decay"},{anchor:"bitsandbytes.optim.AdamW.__init__.amsgrad",description:`<strong>amsgrad</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to use the <a href="https://hf.co/papers/1904.09237" rel="nofollow">AMSGrad</a> variant of Adam that uses the maximum of past squared gradients instead.`,name:"amsgrad"},{anchor:"bitsandbytes.optim.AdamW.__init__.optim_bits",description:`<strong>optim_bits</strong> (<code>int</code>, defaults to 32) &#x2014;
The number of bits of the optimizer state.`,name:"optim_bits"},{anchor:"bitsandbytes.optim.AdamW.__init__.args",description:`<strong>args</strong> (<code>object</code>, defaults to <code>None</code>) &#x2014;
An object with additional arguments.`,name:"args"},{anchor:"bitsandbytes.optim.AdamW.__init__.min_8bit_size",description:`<strong>min_8bit_size</strong> (<code>int</code>, defaults to 4096) &#x2014;
The minimum number of elements of the parameter tensors for 8-bit optimization.`,name:"min_8bit_size"},{anchor:"bitsandbytes.optim.AdamW.__init__.is_paged",description:`<strong>is_paged</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether the optimizer is a paged optimizer or not.`,name:"is_paged"}],source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_1908/bitsandbytes/optim/adamw.py#L10"}}),S=new et({props:{title:"AdamW8bit",local:"bitsandbytes.optim.AdamW8bit",headingTag:"h2"}}),G=new f({props:{name:"class bitsandbytes.optim.AdamW8bit",anchor:"bitsandbytes.optim.AdamW8bit",parameters:[{name:"params",val:""},{name:"lr",val:" = 0.001"},{name:"betas",val:" = (0.9, 0.999)"},{name:"eps",val:" = 1e-08"},{name:"weight_decay",val:" = 0.01"},{name:"amsgrad",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"},{name:"is_paged",val:" = False"}],source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_1908/bitsandbytes/optim/adamw.py#L62"}}),q=new f({props:{name:"__init__",anchor:"bitsandbytes.optim.AdamW8bit.__init__",parameters:[{name:"params",val:""},{name:"lr",val:" = 0.001"},{name:"betas",val:" = (0.9, 0.999)"},{name:"eps",val:" = 1e-08"},{name:"weight_decay",val:" = 0.01"},{name:"amsgrad",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"},{name:"is_paged",val:" = False"}],parametersDescription:[{anchor:"bitsandbytes.optim.AdamW8bit.__init__.params",description:`<strong>params</strong> (<code>torch.Tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.AdamW8bit.__init__.lr",description:`<strong>lr</strong> (<code>float</code>, defaults to 1e-3) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.AdamW8bit.__init__.betas",description:`<strong>betas</strong> (<code>tuple(float, float)</code>, defaults to (0.9, 0.999)) &#x2014;
The beta values are the decay rates of the first and second-order moment of the optimizer.`,name:"betas"},{anchor:"bitsandbytes.optim.AdamW8bit.__init__.eps",description:`<strong>eps</strong> (<code>float</code>, defaults to 1e-8) &#x2014;
The epsilon value prevents division by zero in the optimizer.`,name:"eps"},{anchor:"bitsandbytes.optim.AdamW8bit.__init__.weight_decay",description:`<strong>weight_decay</strong> (<code>float</code>, defaults to 1e-2) &#x2014;
The weight decay value for the optimizer.`,name:"weight_decay"},{anchor:"bitsandbytes.optim.AdamW8bit.__init__.amsgrad",description:`<strong>amsgrad</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to use the <a href="https://hf.co/papers/1904.09237" rel="nofollow">AMSGrad</a> variant of Adam that uses the maximum of past squared gradients instead.
Note: This parameter is not supported in AdamW8bit and must be False.`,name:"amsgrad"},{anchor:"bitsandbytes.optim.AdamW8bit.__init__.optim_bits",description:`<strong>optim_bits</strong> (<code>int</code>, defaults to 32) &#x2014;
The number of bits of the optimizer state.
Note: This parameter is not used in AdamW8bit as it always uses 8-bit optimization.`,name:"optim_bits"},{anchor:"bitsandbytes.optim.AdamW8bit.__init__.args",description:`<strong>args</strong> (<code>object</code>, defaults to <code>None</code>) &#x2014;
An object with additional arguments.`,name:"args"},{anchor:"bitsandbytes.optim.AdamW8bit.__init__.min_8bit_size",description:`<strong>min_8bit_size</strong> (<code>int</code>, defaults to 4096) &#x2014;
The minimum number of elements of the parameter tensors for 8-bit optimization.`,name:"min_8bit_size"},{anchor:"bitsandbytes.optim.AdamW8bit.__init__.is_paged",description:`<strong>is_paged</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether the optimizer is a paged optimizer or not.`,name:"is_paged"}],source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_1908/bitsandbytes/optim/adamw.py#L63"}}),U=new et({props:{title:"AdamW32bit",local:"bitsandbytes.optim.AdamW32bit",headingTag:"h2"}}),H=new f({props:{name:"class bitsandbytes.optim.AdamW32bit",anchor:"bitsandbytes.optim.AdamW32bit",parameters:[{name:"params",val:""},{name:"lr",val:" = 0.001"},{name:"betas",val:" = (0.9, 0.999)"},{name:"eps",val:" = 1e-08"},{name:"weight_decay",val:" = 0.01"},{name:"amsgrad",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"},{name:"is_paged",val:" = False"}],source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_1908/bitsandbytes/optim/adamw.py#L126"}}),k=new f({props:{name:"__init__",anchor:"bitsandbytes.optim.AdamW32bit.__init__",parameters:[{name:"params",val:""},{name:"lr",val:" = 0.001"},{name:"betas",val:" = (0.9, 0.999)"},{name:"eps",val:" = 1e-08"},{name:"weight_decay",val:" = 0.01"},{name:"amsgrad",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"},{name:"is_paged",val:" = False"}],parametersDescription:[{anchor:"bitsandbytes.optim.AdamW32bit.__init__.params",description:`<strong>params</strong> (<code>torch.Tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.AdamW32bit.__init__.lr",description:`<strong>lr</strong> (<code>float</code>, defaults to 1e-3) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.AdamW32bit.__init__.betas",description:`<strong>betas</strong> (<code>tuple(float, float)</code>, defaults to (0.9, 0.999)) &#x2014;
The beta values are the decay rates of the first and second-order moment of the optimizer.`,name:"betas"},{anchor:"bitsandbytes.optim.AdamW32bit.__init__.eps",description:`<strong>eps</strong> (<code>float</code>, defaults to 1e-8) &#x2014;
The epsilon value prevents division by zero in the optimizer.`,name:"eps"},{anchor:"bitsandbytes.optim.AdamW32bit.__init__.weight_decay",description:`<strong>weight_decay</strong> (<code>float</code>, defaults to 1e-2) &#x2014;
The weight decay value for the optimizer.`,name:"weight_decay"},{anchor:"bitsandbytes.optim.AdamW32bit.__init__.amsgrad",description:`<strong>amsgrad</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to use the <a href="https://hf.co/papers/1904.09237" rel="nofollow">AMSGrad</a> variant of Adam that uses the maximum of past squared gradients instead.`,name:"amsgrad"},{anchor:"bitsandbytes.optim.AdamW32bit.__init__.optim_bits",description:`<strong>optim_bits</strong> (<code>int</code>, defaults to 32) &#x2014;
The number of bits of the optimizer state.`,name:"optim_bits"},{anchor:"bitsandbytes.optim.AdamW32bit.__init__.args",description:`<strong>args</strong> (<code>object</code>, defaults to <code>None</code>) &#x2014;
An object with additional arguments.`,name:"args"},{anchor:"bitsandbytes.optim.AdamW32bit.__init__.min_8bit_size",description:`<strong>min_8bit_size</strong> (<code>int</code>, defaults to 4096) &#x2014;
The minimum number of elements of the parameter tensors for 8-bit optimization.`,name:"min_8bit_size"},{anchor:"bitsandbytes.optim.AdamW32bit.__init__.is_paged",description:`<strong>is_paged</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether the optimizer is a paged optimizer or not.`,name:"is_paged"}],source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_1908/bitsandbytes/optim/adamw.py#L127"}}),B=new et({props:{title:"PagedAdamW",local:"bitsandbytes.optim.PagedAdamW",headingTag:"h2"}}),O=new f({props:{name:"class bitsandbytes.optim.PagedAdamW",anchor:"bitsandbytes.optim.PagedAdamW",parameters:[{name:"params",val:""},{name:"lr",val:" = 0.001"},{name:"betas",val:" = (0.9, 0.999)"},{name:"eps",val:" = 1e-08"},{name:"weight_decay",val:" = 0.01"},{name:"amsgrad",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"}],source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_1908/bitsandbytes/optim/adamw.py#L179"}}),R=new f({props:{name:"__init__",anchor:"bitsandbytes.optim.PagedAdamW.__init__",parameters:[{name:"params",val:""},{name:"lr",val:" = 0.001"},{name:"betas",val:" = (0.9, 0.999)"},{name:"eps",val:" = 1e-08"},{name:"weight_decay",val:" = 0.01"},{name:"amsgrad",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"}],parametersDescription:[{anchor:"bitsandbytes.optim.PagedAdamW.__init__.params",description:`<strong>params</strong> (<code>torch.Tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.PagedAdamW.__init__.lr",description:`<strong>lr</strong> (<code>float</code>, defaults to 1e-3) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.PagedAdamW.__init__.betas",description:`<strong>betas</strong> (<code>tuple(float, float)</code>, defaults to (0.9, 0.999)) &#x2014;
The beta values are the decay rates of the first and second-order moment of the optimizer.`,name:"betas"},{anchor:"bitsandbytes.optim.PagedAdamW.__init__.eps",description:`<strong>eps</strong> (<code>float</code>, defaults to 1e-8) &#x2014;
The epsilon value prevents division by zero in the optimizer.`,name:"eps"},{anchor:"bitsandbytes.optim.PagedAdamW.__init__.weight_decay",description:`<strong>weight_decay</strong> (<code>float</code>, defaults to 1e-2) &#x2014;
The weight decay value for the optimizer.`,name:"weight_decay"},{anchor:"bitsandbytes.optim.PagedAdamW.__init__.amsgrad",description:`<strong>amsgrad</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to use the <a href="https://hf.co/papers/1904.09237" rel="nofollow">AMSGrad</a> variant of Adam that uses the maximum of past squared gradients instead.`,name:"amsgrad"},{anchor:"bitsandbytes.optim.PagedAdamW.__init__.optim_bits",description:`<strong>optim_bits</strong> (<code>int</code>, defaults to 32) &#x2014;
The number of bits of the optimizer state.`,name:"optim_bits"},{anchor:"bitsandbytes.optim.PagedAdamW.__init__.args",description:`<strong>args</strong> (<code>object</code>, defaults to <code>None</code>) &#x2014;
An object with additional arguments.`,name:"args"},{anchor:"bitsandbytes.optim.PagedAdamW.__init__.min_8bit_size",description:`<strong>min_8bit_size</strong> (<code>int</code>, defaults to 4096) &#x2014;
The minimum number of elements of the parameter tensors for 8-bit optimization.`,name:"min_8bit_size"}],source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_1908/bitsandbytes/optim/adamw.py#L180"}}),J=new et({props:{title:"PagedAdamW8bit",local:"bitsandbytes.optim.PagedAdamW8bit",headingTag:"h2"}}),K=new f({props:{name:"class bitsandbytes.optim.PagedAdamW8bit",anchor:"bitsandbytes.optim.PagedAdamW8bit",parameters:[{name:"params",val:""},{name:"lr",val:" = 0.001"},{name:"betas",val:" = (0.9, 0.999)"},{name:"eps",val:" = 1e-08"},{name:"weight_decay",val:" = 0.01"},{name:"amsgrad",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"}],source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_1908/bitsandbytes/optim/adamw.py#L229"}}),Q=new f({props:{name:"__init__",anchor:"bitsandbytes.optim.PagedAdamW8bit.__init__",parameters:[{name:"params",val:""},{name:"lr",val:" = 0.001"},{name:"betas",val:" = (0.9, 0.999)"},{name:"eps",val:" = 1e-08"},{name:"weight_decay",val:" = 0.01"},{name:"amsgrad",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"}],parametersDescription:[{anchor:"bitsandbytes.optim.PagedAdamW8bit.__init__.params",description:`<strong>params</strong> (<code>torch.Tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.PagedAdamW8bit.__init__.lr",description:`<strong>lr</strong> (<code>float</code>, defaults to 1e-3) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.PagedAdamW8bit.__init__.betas",description:`<strong>betas</strong> (<code>tuple(float, float)</code>, defaults to (0.9, 0.999)) &#x2014;
The beta values are the decay rates of the first and second-order moment of the optimizer.`,name:"betas"},{anchor:"bitsandbytes.optim.PagedAdamW8bit.__init__.eps",description:`<strong>eps</strong> (<code>float</code>, defaults to 1e-8) &#x2014;
The epsilon value prevents division by zero in the optimizer.`,name:"eps"},{anchor:"bitsandbytes.optim.PagedAdamW8bit.__init__.weight_decay",description:`<strong>weight_decay</strong> (<code>float</code>, defaults to 1e-2) &#x2014;
The weight decay value for the optimizer.`,name:"weight_decay"},{anchor:"bitsandbytes.optim.PagedAdamW8bit.__init__.amsgrad",description:`<strong>amsgrad</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to use the <a href="https://hf.co/papers/1904.09237" rel="nofollow">AMSGrad</a> variant of Adam that uses the maximum of past squared gradients instead.
Note: This parameter is not supported in PagedAdamW8bit and must be False.`,name:"amsgrad"},{anchor:"bitsandbytes.optim.PagedAdamW8bit.__init__.optim_bits",description:`<strong>optim_bits</strong> (<code>int</code>, defaults to 32) &#x2014;
The number of bits of the optimizer state.
Note: This parameter is not used in PagedAdamW8bit as it always uses 8-bit optimization.`,name:"optim_bits"},{anchor:"bitsandbytes.optim.PagedAdamW8bit.__init__.args",description:`<strong>args</strong> (<code>object</code>, defaults to <code>None</code>) &#x2014;
An object with additional arguments.`,name:"args"},{anchor:"bitsandbytes.optim.PagedAdamW8bit.__init__.min_8bit_size",description:`<strong>min_8bit_size</strong> (<code>int</code>, defaults to 4096) &#x2014;
The minimum number of elements of the parameter tensors for 8-bit optimization.`,name:"min_8bit_size"}],source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_1908/bitsandbytes/optim/adamw.py#L230"}}),X=new et({props:{title:"PagedAdamW32bit",local:"bitsandbytes.optim.PagedAdamW32bit",headingTag:"h2"}}),Y=new f({props:{name:"class bitsandbytes.optim.PagedAdamW32bit",anchor:"bitsandbytes.optim.PagedAdamW32bit",parameters:[{name:"params",val:""},{name:"lr",val:" = 0.001"},{name:"betas",val:" = (0.9, 0.999)"},{name:"eps",val:" = 1e-08"},{name:"weight_decay",val:" = 0.01"},{name:"amsgrad",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"}],source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_1908/bitsandbytes/optim/adamw.py#L290"}}),Z=new f({props:{name:"__init__",anchor:"bitsandbytes.optim.PagedAdamW32bit.__init__",parameters:[{name:"params",val:""},{name:"lr",val:" = 0.001"},{name:"betas",val:" = (0.9, 0.999)"},{name:"eps",val:" = 1e-08"},{name:"weight_decay",val:" = 0.01"},{name:"amsgrad",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"}],parametersDescription:[{anchor:"bitsandbytes.optim.PagedAdamW32bit.__init__.params",description:`<strong>params</strong> (<code>torch.Tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.PagedAdamW32bit.__init__.lr",description:`<strong>lr</strong> (<code>float</code>, defaults to 1e-3) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.PagedAdamW32bit.__init__.betas",description:`<strong>betas</strong> (<code>tuple(float, float)</code>, defaults to (0.9, 0.999)) &#x2014;
The beta values are the decay rates of the first and second-order moment of the optimizer.`,name:"betas"},{anchor:"bitsandbytes.optim.PagedAdamW32bit.__init__.eps",description:`<strong>eps</strong> (<code>float</code>, defaults to 1e-8) &#x2014;
The epsilon value prevents division by zero in the optimizer.`,name:"eps"},{anchor:"bitsandbytes.optim.PagedAdamW32bit.__init__.weight_decay",description:`<strong>weight_decay</strong> (<code>float</code>, defaults to 1e-2) &#x2014;
The weight decay value for the optimizer.`,name:"weight_decay"},{anchor:"bitsandbytes.optim.PagedAdamW32bit.__init__.amsgrad",description:`<strong>amsgrad</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to use the <a href="https://hf.co/papers/1904.09237" rel="nofollow">AMSGrad</a> variant of Adam that uses the maximum of past squared gradients instead.`,name:"amsgrad"},{anchor:"bitsandbytes.optim.PagedAdamW32bit.__init__.optim_bits",description:`<strong>optim_bits</strong> (<code>int</code>, defaults to 32) &#x2014;
The number of bits of the optimizer state.`,name:"optim_bits"},{anchor:"bitsandbytes.optim.PagedAdamW32bit.__init__.args",description:`<strong>args</strong> (<code>object</code>, defaults to <code>None</code>) &#x2014;
An object with additional arguments.`,name:"args"},{anchor:"bitsandbytes.optim.PagedAdamW32bit.__init__.min_8bit_size",description:`<strong>min_8bit_size</strong> (<code>int</code>, defaults to 4096) &#x2014;
The minimum number of elements of the parameter tensors for 8-bit optimization.`,name:"min_8bit_size"}],source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_1908/bitsandbytes/optim/adamw.py#L291"}}),tt=new he({props:{source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/main/docs/source/reference/optim/adamw.mdx"}}),{c(){u=r("meta"),bt=i(),dt=r("p"),pt=i(),m(N.$$.fragment),lt=i(),m(E.$$.fragment),_t=i(),L=r("p"),L.innerHTML=ee,ct=i(),j=r("p"),j.textContent=ae,gt=i(),m(I.$$.fragment),ht=i(),v=r("div"),m(M.$$.fragment),Ut=i(),w=r("div"),m(V.$$.fragment),Ht=i(),at=r("p"),at.textContent=ie,ft=i(),m(S.$$.fragment),ut=i(),y=r("div"),m(G.$$.fragment),kt=i(),z=r("div"),m(q.$$.fragment),Bt=i(),it=r("p"),it.textContent=ne,vt=i(),m(U.$$.fragment),yt=i(),$=r("div"),m(H.$$.fragment),Ot=i(),P=r("div"),m(k.$$.fragment),Rt=i(),nt=r("p"),nt.textContent=se,$t=i(),m(B.$$.fragment),At=i(),A=r("div"),m(O.$$.fragment),Jt=i(),T=r("div"),m(R.$$.fragment),Kt=i(),st=r("p"),st.textContent=oe,Wt=i(),m(J.$$.fragment),xt=i(),W=r("div"),m(K.$$.fragment),Qt=i(),F=r("div"),m(Q.$$.fragment),Xt=i(),ot=r("p"),ot.textContent=re,wt=i(),m(X.$$.fragment),zt=i(),x=r("div"),m(Y.$$.fragment),Yt=i(),C=r("div"),m(Z.$$.fragment),Zt=i(),rt=r("p"),rt.textContent=de,Pt=i(),m(tt.$$.fragment),Tt=i(),mt=r("p"),this.h()},l(t){const a=ce("svelte-u9bgzb",document.head);u=d(a,"META",{name:!0,content:!0}),a.forEach(e),bt=n(t),dt=d(t,"P",{}),g(dt).forEach(e),pt=n(t),b(N.$$.fragment,t),lt=n(t),b(E.$$.fragment,t),_t=n(t),L=d(t,"P",{"data-svelte-h":!0}),D(L)!=="svelte-zypq46"&&(L.innerHTML=ee),ct=n(t),j=d(t,"P",{"data-svelte-h":!0}),D(j)!=="svelte-qpasov"&&(j.textContent=ae),gt=n(t),b(I.$$.fragment,t),ht=n(t),v=d(t,"DIV",{class:!0});var Ct=g(v);b(M.$$.fragment,Ct),Ut=n(Ct),w=d(Ct,"DIV",{class:!0});var Dt=g(w);b(V.$$.fragment,Dt),Ht=n(Dt),at=d(Dt,"P",{"data-svelte-h":!0}),D(at)!=="svelte-cz2sic"&&(at.textContent=ie),Dt.forEach(e),Ct.forEach(e),ft=n(t),b(S.$$.fragment,t),ut=n(t),y=d(t,"DIV",{class:!0});var Nt=g(y);b(G.$$.fragment,Nt),kt=n(Nt),z=d(Nt,"DIV",{class:!0});var Et=g(z);b(q.$$.fragment,Et),Bt=n(Et),it=d(Et,"P",{"data-svelte-h":!0}),D(it)!=="svelte-1jiw1x7"&&(it.textContent=ne),Et.forEach(e),Nt.forEach(e),vt=n(t),b(U.$$.fragment,t),yt=n(t),$=d(t,"DIV",{class:!0});var Lt=g($);b(H.$$.fragment,Lt),Ot=n(Lt),P=d(Lt,"DIV",{class:!0});var jt=g(P);b(k.$$.fragment,jt),Rt=n(jt),nt=d(jt,"P",{"data-svelte-h":!0}),D(nt)!=="svelte-gm8nxe"&&(nt.textContent=se),jt.forEach(e),Lt.forEach(e),$t=n(t),b(B.$$.fragment,t),At=n(t),A=d(t,"DIV",{class:!0});var It=g(A);b(O.$$.fragment,It),Jt=n(It),T=d(It,"DIV",{class:!0});var Mt=g(T);b(R.$$.fragment,Mt),Kt=n(Mt),st=d(Mt,"P",{"data-svelte-h":!0}),D(st)!=="svelte-we884g"&&(st.textContent=oe),Mt.forEach(e),It.forEach(e),Wt=n(t),b(J.$$.fragment,t),xt=n(t),W=d(t,"DIV",{class:!0});var Vt=g(W);b(K.$$.fragment,Vt),Qt=n(Vt),F=d(Vt,"DIV",{class:!0});var St=g(F);b(Q.$$.fragment,St),Xt=n(St),ot=d(St,"P",{"data-svelte-h":!0}),D(ot)!=="svelte-y5nwi"&&(ot.textContent=re),St.forEach(e),Vt.forEach(e),wt=n(t),b(X.$$.fragment,t),zt=n(t),x=d(t,"DIV",{class:!0});var Gt=g(x);b(Y.$$.fragment,Gt),Yt=n(Gt),C=d(Gt,"DIV",{class:!0});var qt=g(C);b(Z.$$.fragment,qt),Zt=n(qt),rt=d(qt,"P",{"data-svelte-h":!0}),D(rt)!=="svelte-10llmb"&&(rt.textContent=de),qt.forEach(e),Gt.forEach(e),Pt=n(t),b(tt.$$.fragment,t),Tt=n(t),mt=d(t,"P",{}),g(mt).forEach(e),this.h()},h(){h(u,"name","hf:doc:metadata"),h(u,"content",ue),h(w,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(v,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(z,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(y,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(P,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h($,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(T,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(A,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(F,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(W,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(C,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(x,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8")},m(t,a){o(document.head,u),s(t,bt,a),s(t,dt,a),s(t,pt,a),p(N,t,a),s(t,lt,a),p(E,t,a),s(t,_t,a),s(t,L,a),s(t,ct,a),s(t,j,a),s(t,gt,a),p(I,t,a),s(t,ht,a),s(t,v,a),p(M,v,null),o(v,Ut),o(v,w),p(V,w,null),o(w,Ht),o(w,at),s(t,ft,a),p(S,t,a),s(t,ut,a),s(t,y,a),p(G,y,null),o(y,kt),o(y,z),p(q,z,null),o(z,Bt),o(z,it),s(t,vt,a),p(U,t,a),s(t,yt,a),s(t,$,a),p(H,$,null),o($,Ot),o($,P),p(k,P,null),o(P,Rt),o(P,nt),s(t,$t,a),p(B,t,a),s(t,At,a),s(t,A,a),p(O,A,null),o(A,Jt),o(A,T),p(R,T,null),o(T,Kt),o(T,st),s(t,Wt,a),p(J,t,a),s(t,xt,a),s(t,W,a),p(K,W,null),o(W,Qt),o(W,F),p(Q,F,null),o(F,Xt),o(F,ot),s(t,wt,a),p(X,t,a),s(t,zt,a),s(t,x,a),p(Y,x,null),o(x,Yt),o(x,C),p(Z,C,null),o(C,Zt),o(C,rt),s(t,Pt,a),p(tt,t,a),s(t,Tt,a),s(t,mt,a),Ft=!0},p:be,i(t){Ft||(l(N.$$.fragment,t),l(E.$$.fragment,t),l(I.$$.fragment,t),l(M.$$.fragment,t),l(V.$$.fragment,t),l(S.$$.fragment,t),l(G.$$.fragment,t),l(q.$$.fragment,t),l(U.$$.fragment,t),l(H.$$.fragment,t),l(k.$$.fragment,t),l(B.$$.fragment,t),l(O.$$.fragment,t),l(R.$$.fragment,t),l(J.$$.fragment,t),l(K.$$.fragment,t),l(Q.$$.fragment,t),l(X.$$.fragment,t),l(Y.$$.fragment,t),l(Z.$$.fragment,t),l(tt.$$.fragment,t),Ft=!0)},o(t){_(N.$$.fragment,t),_(E.$$.fragment,t),_(I.$$.fragment,t),_(M.$$.fragment,t),_(V.$$.fragment,t),_(S.$$.fragment,t),_(G.$$.fragment,t),_(q.$$.fragment,t),_(U.$$.fragment,t),_(H.$$.fragment,t),_(k.$$.fragment,t),_(B.$$.fragment,t),_(O.$$.fragment,t),_(R.$$.fragment,t),_(J.$$.fragment,t),_(K.$$.fragment,t),_(Q.$$.fragment,t),_(X.$$.fragment,t),_(Y.$$.fragment,t),_(Z.$$.fragment,t),_(tt.$$.fragment,t),Ft=!1},d(t){t&&(e(bt),e(dt),e(pt),e(lt),e(_t),e(L),e(ct),e(j),e(gt),e(ht),e(v),e(ft),e(ut),e(y),e(vt),e(yt),e($),e($t),e(At),e(A),e(Wt),e(xt),e(W),e(wt),e(zt),e(x),e(Pt),e(Tt),e(mt)),e(u),c(N,t),c(E,t),c(I,t),c(M),c(V),c(S,t),c(G),c(q),c(U,t),c(H),c(k),c(B,t),c(O),c(R),c(J,t),c(K),c(Q),c(X,t),c(Y),c(Z),c(tt,t)}}}const ue='{"title":"AdamW","local":"adamw","sections":[{"title":"AdamW","local":"api-class ][ bitsandbytes.optim.AdamW","sections":[],"depth":2},{"title":"AdamW8bit","local":"bitsandbytes.optim.AdamW8bit","sections":[],"depth":2},{"title":"AdamW32bit","local":"bitsandbytes.optim.AdamW32bit","sections":[],"depth":2},{"title":"PagedAdamW","local":"bitsandbytes.optim.PagedAdamW","sections":[],"depth":2},{"title":"PagedAdamW8bit","local":"bitsandbytes.optim.PagedAdamW8bit","sections":[],"depth":2},{"title":"PagedAdamW32bit","local":"bitsandbytes.optim.PagedAdamW32bit","sections":[],"depth":2}],"depth":1}';function ve(te){return pe(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class xe extends le{constructor(u){super(),_e(this,u,ve,fe,me,{})}}export{xe as component};