bitsandbytes/pr_1925/en/_app/immutable/nodes/18.c5d5f84e.js

import{s as le,n as _e,o as ge}from"../chunks/scheduler.8a2cc2fa.js";import{S as ce,i as he,e as r,s as i,c as m,h as fe,a as d,d as e,b as n,f as c,g as p,j as w,k as h,l as o,m as s,n as b,t as l,o as _,p as g}from"../chunks/index.7079e750.js";import{C as ue,H as at,E as ve}from"../chunks/MermaidChart.svelte_svelte_type_style_lang.0d032bd2.js";import{D as f}from"../chunks/Docstring.a1d84405.js";function ye(ae){let u,bt,mt,lt,L,_t,E,gt,M,ie='<a href="https://hf.co/papers/1412.6980" rel="nofollow">Adam (Adaptive moment estimation)</a> is an adaptive learning rate optimizer, combining ideas from <code>SGD</code> with momentum and <code>RMSprop</code> to automatically scale the learning rate:',ct,j,ne="<li>a weighted average of the past gradients to provide direction (first-moment)</li> <li>a weighted average of the <em>squared</em> past gradients to adapt the learning rate to each parameter (second-moment)</li>",ht,I,se="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.",ft,S,ut,v,V,kt,P,W,Rt,it,oe="Base Adam optimizer.",vt,q,yt,y,G,Bt,T,H,Ot,nt,re="8-bit Adam optimizer.",$t,U,At,$,k,Jt,F,R,Kt,st,de="32-bit Adam optimizer.",xt,B,zt,A,O,Qt,C,J,Xt,ot,me="Paged Adam optimizer.",wt,K,Pt,x,Q,Yt,D,X,Zt,rt,pe="8-bit paged Adam optimizer.",Tt,Y,Ft,z,Z,te,N,tt,ee,dt,be="Paged 32-bit Adam optimizer.",Ct,et,Dt,pt,Nt;return L=new ue({props:{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"}}),E=new at({props:{title:"Adam",local:"adam",headingTag:"h1"}}),S=new at({props:{title:"Adam",local:"api-class ][ bitsandbytes.optim.Adam",headingTag:"h2"}}),V=new f({props:{name:"class bitsandbytes.optim.Adam",anchor:"bitsandbytes.optim.Adam",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"},{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_1925/bitsandbytes/optim/adam.py#L9"}}),W=new f({props:{name:"__init__",anchor:"bitsandbytes.optim.Adam.__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"},{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.Adam.__init__.params",description:`<strong>params</strong> (<code>torch.tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.Adam.__init__.lr",description:`<strong>lr</strong> (<code>float</code>, defaults to 1e-3) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.Adam.__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.Adam.__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.Adam.__init__.weight_decay",description:`<strong>weight_decay</strong> (<code>float</code>, defaults to 0.0) &#x2014;
The weight decay value for the optimizer.`,name:"weight_decay"},{anchor:"bitsandbytes.optim.Adam.__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.Adam.__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.Adam.__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.Adam.__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.Adam.__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_1925/bitsandbytes/optim/adam.py#L10"}}),q=new at({props:{title:"Adam8bit",local:"bitsandbytes.optim.Adam8bit",headingTag:"h2"}}),G=new f({props:{name:"class bitsandbytes.optim.Adam8bit",anchor:"bitsandbytes.optim.Adam8bit",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"},{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_1925/bitsandbytes/optim/adam.py#L62"}}),H=new f({props:{name:"__init__",anchor:"bitsandbytes.optim.Adam8bit.__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"},{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.Adam8bit.__init__.params",description:`<strong>params</strong> (<code>torch.tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.Adam8bit.__init__.lr",description:`<strong>lr</strong> (<code>float</code>, defaults to 1e-3) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.Adam8bit.__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.Adam8bit.__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.Adam8bit.__init__.weight_decay",description:`<strong>weight_decay</strong> (<code>float</code>, defaults to 0.0) &#x2014;
The weight decay value for the optimizer.`,name:"weight_decay"},{anchor:"bitsandbytes.optim.Adam8bit.__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 Adam8bit and must be False.`,name:"amsgrad"},{anchor:"bitsandbytes.optim.Adam8bit.__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 Adam8bit as it always uses 8-bit optimization.`,name:"optim_bits"},{anchor:"bitsandbytes.optim.Adam8bit.__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.Adam8bit.__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.Adam8bit.__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_1925/bitsandbytes/optim/adam.py#L63"}}),U=new at({props:{title:"Adam32bit",local:"bitsandbytes.optim.Adam32bit",headingTag:"h2"}}),k=new f({props:{name:"class bitsandbytes.optim.Adam32bit",anchor:"bitsandbytes.optim.Adam32bit",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"},{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_1925/bitsandbytes/optim/adam.py#L126"}}),R=new f({props:{name:"__init__",anchor:"bitsandbytes.optim.Adam32bit.__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"},{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.Adam32bit.__init__.params",description:`<strong>params</strong> (<code>torch.tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.Adam32bit.__init__.lr",description:`<strong>lr</strong> (<code>float</code>, defaults to 1e-3) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.Adam32bit.__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.Adam32bit.__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.Adam32bit.__init__.weight_decay",description:`<strong>weight_decay</strong> (<code>float</code>, defaults to 0.0) &#x2014;
The weight decay value for the optimizer.`,name:"weight_decay"},{anchor:"bitsandbytes.optim.Adam32bit.__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.Adam32bit.__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.Adam32bit.__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.Adam32bit.__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.Adam32bit.__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_1925/bitsandbytes/optim/adam.py#L127"}}),B=new at({props:{title:"PagedAdam",local:"bitsandbytes.optim.PagedAdam",headingTag:"h2"}}),O=new f({props:{name:"class bitsandbytes.optim.PagedAdam",anchor:"bitsandbytes.optim.PagedAdam",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"},{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_1925/bitsandbytes/optim/adam.py#L179"}}),J=new f({props:{name:"__init__",anchor:"bitsandbytes.optim.PagedAdam.__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"},{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.PagedAdam.__init__.params",description:`<strong>params</strong> (<code>torch.tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.PagedAdam.__init__.lr",description:`<strong>lr</strong> (<code>float</code>, defaults to 1e-3) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.PagedAdam.__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.PagedAdam.__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.PagedAdam.__init__.weight_decay",description:`<strong>weight_decay</strong> (<code>float</code>, defaults to 0.0) &#x2014;
The weight decay value for the optimizer.`,name:"weight_decay"},{anchor:"bitsandbytes.optim.PagedAdam.__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.PagedAdam.__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.PagedAdam.__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.PagedAdam.__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.PagedAdam.__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_1925/bitsandbytes/optim/adam.py#L180"}}),K=new at({props:{title:"PagedAdam8bit",local:"bitsandbytes.optim.PagedAdam8bit",headingTag:"h2"}}),Q=new f({props:{name:"class bitsandbytes.optim.PagedAdam8bit",anchor:"bitsandbytes.optim.PagedAdam8bit",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"},{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_1925/bitsandbytes/optim/adam.py#L232"}}),X=new f({props:{name:"__init__",anchor:"bitsandbytes.optim.PagedAdam8bit.__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"},{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.PagedAdam8bit.__init__.params",description:`<strong>params</strong> (<code>torch.tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.PagedAdam8bit.__init__.lr",description:`<strong>lr</strong> (<code>float</code>, defaults to 1e-3) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.PagedAdam8bit.__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.PagedAdam8bit.__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.PagedAdam8bit.__init__.weight_decay",description:`<strong>weight_decay</strong> (<code>float</code>, defaults to 0.0) &#x2014;
The weight decay value for the optimizer.`,name:"weight_decay"},{anchor:"bitsandbytes.optim.PagedAdam8bit.__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 PagedAdam8bit and must be False.`,name:"amsgrad"},{anchor:"bitsandbytes.optim.PagedAdam8bit.__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 PagedAdam8bit as it always uses 8-bit optimization.`,name:"optim_bits"},{anchor:"bitsandbytes.optim.PagedAdam8bit.__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.PagedAdam8bit.__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.PagedAdam8bit.__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_1925/bitsandbytes/optim/adam.py#L233"}}),Y=new at({props:{title:"PagedAdam32bit",local:"bitsandbytes.optim.PagedAdam32bit",headingTag:"h2"}}),Z=new f({props:{name:"class bitsandbytes.optim.PagedAdam32bit",anchor:"bitsandbytes.optim.PagedAdam32bit",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"},{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_1925/bitsandbytes/optim/adam.py#L296"}}),tt=new f({props:{name:"__init__",anchor:"bitsandbytes.optim.PagedAdam32bit.__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"},{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.PagedAdam32bit.__init__.params",description:`<strong>params</strong> (<code>torch.tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.PagedAdam32bit.__init__.lr",description:`<strong>lr</strong> (<code>float</code>, defaults to 1e-3) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.PagedAdam32bit.__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.PagedAdam32bit.__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.PagedAdam32bit.__init__.weight_decay",description:`<strong>weight_decay</strong> (<code>float</code>, defaults to 0.0) &#x2014;
The weight decay value for the optimizer.`,name:"weight_decay"},{anchor:"bitsandbytes.optim.PagedAdam32bit.__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.PagedAdam32bit.__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.PagedAdam32bit.__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.PagedAdam32bit.__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.PagedAdam32bit.__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_1925/bitsandbytes/optim/adam.py#L297"}}),et=new ve({props:{source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/main/docs/source/reference/optim/adam.mdx"}}),{c(){u=r("meta"),bt=i(),mt=r("p"),lt=i(),m(L.$$.fragment),_t=i(),m(E.$$.fragment),gt=i(),M=r("p"),M.innerHTML=ie,ct=i(),j=r("ul"),j.innerHTML=ne,ht=i(),I=r("p"),I.textContent=se,ft=i(),m(S.$$.fragment),ut=i(),v=r("div"),m(V.$$.fragment),kt=i(),P=r("div"),m(W.$$.fragment),Rt=i(),it=r("p"),it.textContent=oe,vt=i(),m(q.$$.fragment),yt=i(),y=r("div"),m(G.$$.fragment),Bt=i(),T=r("div"),m(H.$$.fragment),Ot=i(),nt=r("p"),nt.textContent=re,$t=i(),m(U.$$.fragment),At=i(),$=r("div"),m(k.$$.fragment),Jt=i(),F=r("div"),m(R.$$.fragment),Kt=i(),st=r("p"),st.textContent=de,xt=i(),m(B.$$.fragment),zt=i(),A=r("div"),m(O.$$.fragment),Qt=i(),C=r("div"),m(J.$$.fragment),Xt=i(),ot=r("p"),ot.textContent=me,wt=i(),m(K.$$.fragment),Pt=i(),x=r("div"),m(Q.$$.fragment),Yt=i(),D=r("div"),m(X.$$.fragment),Zt=i(),rt=r("p"),rt.textContent=pe,Tt=i(),m(Y.$$.fragment),Ft=i(),z=r("div"),m(Z.$$.fragment),te=i(),N=r("div"),m(tt.$$.fragment),ee=i(),dt=r("p"),dt.textContent=be,Ct=i(),m(et.$$.fragment),Dt=i(),pt=r("p"),this.h()},l(t){const a=fe("svelte-u9bgzb",document.head);u=d(a,"META",{name:!0,content:!0}),a.forEach(e),bt=n(t),mt=d(t,"P",{}),c(mt).forEach(e),lt=n(t),p(L.$$.fragment,t),_t=n(t),p(E.$$.fragment,t),gt=n(t),M=d(t,"P",{"data-svelte-h":!0}),w(M)!=="svelte-et5h9y"&&(M.innerHTML=ie),ct=n(t),j=d(t,"UL",{"data-svelte-h":!0}),w(j)!=="svelte-1miac0o"&&(j.innerHTML=ne),ht=n(t),I=d(t,"P",{"data-svelte-h":!0}),w(I)!=="svelte-qpasov"&&(I.textContent=se),ft=n(t),p(S.$$.fragment,t),ut=n(t),v=d(t,"DIV",{class:!0});var Lt=c(v);p(V.$$.fragment,Lt),kt=n(Lt),P=d(Lt,"DIV",{class:!0});var Et=c(P);p(W.$$.fragment,Et),Rt=n(Et),it=d(Et,"P",{"data-svelte-h":!0}),w(it)!=="svelte-o3b8f9"&&(it.textContent=oe),Et.forEach(e),Lt.forEach(e),vt=n(t),p(q.$$.fragment,t),yt=n(t),y=d(t,"DIV",{class:!0});var Mt=c(y);p(G.$$.fragment,Mt),Bt=n(Mt),T=d(Mt,"DIV",{class:!0});var jt=c(T);p(H.$$.fragment,jt),Ot=n(jt),nt=d(jt,"P",{"data-svelte-h":!0}),w(nt)!=="svelte-17h2o0s"&&(nt.textContent=re),jt.forEach(e),Mt.forEach(e),$t=n(t),p(U.$$.fragment,t),At=n(t),$=d(t,"DIV",{class:!0});var It=c($);p(k.$$.fragment,It),Jt=n(It),F=d(It,"DIV",{class:!0});var St=c(F);p(R.$$.fragment,St),Kt=n(St),st=d(St,"P",{"data-svelte-h":!0}),w(st)!=="svelte-1x8tgev"&&(st.textContent=de),St.forEach(e),It.forEach(e),xt=n(t),p(B.$$.fragment,t),zt=n(t),A=d(t,"DIV",{class:!0});var Vt=c(A);p(O.$$.fragment,Vt),Qt=n(Vt),C=d(Vt,"DIV",{class:!0});var Wt=c(C);p(J.$$.fragment,Wt),Xt=n(Wt),ot=d(Wt,"P",{"data-svelte-h":!0}),w(ot)!=="svelte-1a1nn0n"&&(ot.textContent=me),Wt.forEach(e),Vt.forEach(e),wt=n(t),p(K.$$.fragment,t),Pt=n(t),x=d(t,"DIV",{class:!0});var qt=c(x);p(Q.$$.fragment,qt),Yt=n(qt),D=d(qt,"DIV",{class:!0});var Gt=c(D);p(X.$$.fragment,Gt),Zt=n(Gt),rt=d(Gt,"P",{"data-svelte-h":!0}),w(rt)!=="svelte-12q2iox"&&(rt.textContent=pe),Gt.forEach(e),qt.forEach(e),Tt=n(t),p(Y.$$.fragment,t),Ft=n(t),z=d(t,"DIV",{class:!0});var Ht=c(z);p(Z.$$.fragment,Ht),te=n(Ht),N=d(Ht,"DIV",{class:!0});var Ut=c(N);p(tt.$$.fragment,Ut),ee=n(Ut),dt=d(Ut,"P",{"data-svelte-h":!0}),w(dt)!=="svelte-1qxzzie"&&(dt.textContent=be),Ut.forEach(e),Ht.forEach(e),Ct=n(t),p(et.$$.fragment,t),Dt=n(t),pt=d(t,"P",{}),c(pt).forEach(e),this.h()},h(){h(u,"name","hf:doc:metadata"),h(u,"content",$e),h(P,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(v,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(T,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(y,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h(F,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),h($,"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(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(D,"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"),h(N,"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")},m(t,a){o(document.head,u),s(t,bt,a),s(t,mt,a),s(t,lt,a),b(L,t,a),s(t,_t,a),b(E,t,a),s(t,gt,a),s(t,M,a),s(t,ct,a),s(t,j,a),s(t,ht,a),s(t,I,a),s(t,ft,a),b(S,t,a),s(t,ut,a),s(t,v,a),b(V,v,null),o(v,kt),o(v,P),b(W,P,null),o(P,Rt),o(P,it),s(t,vt,a),b(q,t,a),s(t,yt,a),s(t,y,a),b(G,y,null),o(y,Bt),o(y,T),b(H,T,null),o(T,Ot),o(T,nt),s(t,$t,a),b(U,t,a),s(t,At,a),s(t,$,a),b(k,$,null),o($,Jt),o($,F),b(R,F,null),o(F,Kt),o(F,st),s(t,xt,a),b(B,t,a),s(t,zt,a),s(t,A,a),b(O,A,null),o(A,Qt),o(A,C),b(J,C,null),o(C,Xt),o(C,ot),s(t,wt,a),b(K,t,a),s(t,Pt,a),s(t,x,a),b(Q,x,null),o(x,Yt),o(x,D),b(X,D,null),o(D,Zt),o(D,rt),s(t,Tt,a),b(Y,t,a),s(t,Ft,a),s(t,z,a),b(Z,z,null),o(z,te),o(z,N),b(tt,N,null),o(N,ee),o(N,dt),s(t,Ct,a),b(et,t,a),s(t,Dt,a),s(t,pt,a),Nt=!0},p:_e,i(t){Nt||(l(L.$$.fragment,t),l(E.$$.fragment,t),l(S.$$.fragment,t),l(V.$$.fragment,t),l(W.$$.fragment,t),l(q.$$.fragment,t),l(G.$$.fragment,t),l(H.$$.fragment,t),l(U.$$.fragment,t),l(k.$$.fragment,t),l(R.$$.fragment,t),l(B.$$.fragment,t),l(O.$$.fragment,t),l(J.$$.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),l(et.$$.fragment,t),Nt=!0)},o(t){_(L.$$.fragment,t),_(E.$$.fragment,t),_(S.$$.fragment,t),_(V.$$.fragment,t),_(W.$$.fragment,t),_(q.$$.fragment,t),_(G.$$.fragment,t),_(H.$$.fragment,t),_(U.$$.fragment,t),_(k.$$.fragment,t),_(R.$$.fragment,t),_(B.$$.fragment,t),_(O.$$.fragment,t),_(J.$$.fragment,t),_(K.$$.fragment,t),_(Q.$$.fragment,t),_(X.$$.fragment,t),_(Y.$$.fragment,t),_(Z.$$.fragment,t),_(tt.$$.fragment,t),_(et.$$.fragment,t),Nt=!1},d(t){t&&(e(bt),e(mt),e(lt),e(_t),e(gt),e(M),e(ct),e(j),e(ht),e(I),e(ft),e(ut),e(v),e(vt),e(yt),e(y),e($t),e(At),e($),e(xt),e(zt),e(A),e(wt),e(Pt),e(x),e(Tt),e(Ft),e(z),e(Ct),e(Dt),e(pt)),e(u),g(L,t),g(E,t),g(S,t),g(V),g(W),g(q,t),g(G),g(H),g(U,t),g(k),g(R),g(B,t),g(O),g(J),g(K,t),g(Q),g(X),g(Y,t),g(Z),g(tt),g(et,t)}}}const $e='{"title":"Adam","local":"adam","sections":[{"title":"Adam","local":"api-class ][ bitsandbytes.optim.Adam","sections":[],"depth":2},{"title":"Adam8bit","local":"bitsandbytes.optim.Adam8bit","sections":[],"depth":2},{"title":"Adam32bit","local":"bitsandbytes.optim.Adam32bit","sections":[],"depth":2},{"title":"PagedAdam","local":"bitsandbytes.optim.PagedAdam","sections":[],"depth":2},{"title":"PagedAdam8bit","local":"bitsandbytes.optim.PagedAdam8bit","sections":[],"depth":2},{"title":"PagedAdam32bit","local":"bitsandbytes.optim.PagedAdam32bit","sections":[],"depth":2}],"depth":1}';function Ae(ae){return ge(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class Te extends ce{constructor(u){super(),he(this,u,Ae,ye,le,{})}}export{Te as component};