bitsandbytes/pr_2038/en/_app/immutable/nodes/16.BciXsF8Z.js

import"../chunks/DsnmJJEf.js";import{i as j,h as k,C as A,H as i,D as e,E as M,s as P}from"../chunks/DdsaYIhL.js";import{p as W,o as q,s as t,f as C,a as G,b as E,c as a,d as x,n as o,r as n}from"../chunks/Cq6T2TQD.js";const H='{"title":"SGD","local":"sgd","sections":[{"title":"SGD","local":"api-class ][ bitsandbytes.optim.SGD","sections":[],"depth":2},{"title":"SGD8bit","local":"bitsandbytes.optim.SGD8bit","sections":[],"depth":2},{"title":"SGD32bit","local":"bitsandbytes.optim.SGD32bit","sections":[],"depth":2}],"depth":1}';var B=x('<meta name="hf:doc:metadata"/>'),O=x('<p></p> <!> <!> <p>Stochastic gradient descent (SGD) is a basic gradient descent optimizer to minimize loss given a set of model parameters and updates the parameters in the opposite direction of the gradient. The update is performed on a randomly sampled mini-batch of data from the dataset.</p> <p>bitsandbytes also supports momentum and Nesterov momentum to accelerate SGD by adding a weighted average of past gradients to the current gradient.</p> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Base SGD optimizer.</p></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>8-bit SGD optimizer.</p></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>32-bit SGD optimizer.</p></div></div> <!> <p></p>',1);function J(z,w){W(w,!1),q(()=>{new URLSearchParams(window.location.search).get("fw")}),j();var d=O();k("1q0mquj",D=>{var S=B();P(S,"content",H),G(D,S)});var b=t(C(d),2);A(b,{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"});var p=t(b,2);i(p,{title:"SGD",local:"sgd",headingTag:"h1"});var c=t(p,6);i(c,{title:"SGD",local:"api-class ][ bitsandbytes.optim.SGD",headingTag:"h2"});var s=t(c,2),l=a(s);e(l,{name:"class bitsandbytes.optim.SGD",anchor:"bitsandbytes.optim.SGD",source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_2038/bitsandbytes/optim/sgd.py#L8",parameters:[{name:"params",val:""},{name:"lr",val:""},{name:"momentum",val:" = 0"},{name:"dampening",val:" = 0"},{name:"weight_decay",val:" = 0"},{name:"nesterov",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"}]});var _=t(l,2),T=a(_);e(T,{name:"__init__",anchor:"bitsandbytes.optim.SGD.__init__",source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_2038/bitsandbytes/optim/sgd.py#L9",parameters:[{name:"params",val:""},{name:"lr",val:""},{name:"momentum",val:" = 0"},{name:"dampening",val:" = 0"},{name:"weight_decay",val:" = 0"},{name:"nesterov",val:" = False"},{name:"optim_bits",val:" = 32"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"}],parametersDescription:[{anchor:"bitsandbytes.optim.SGD.__init__.params",description:`<strong>params</strong> (<code>torch.tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.SGD.__init__.lr",description:`<strong>lr</strong> (<code>float</code>) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.SGD.__init__.momentum",description:`<strong>momentum</strong> (<code>float</code>, defaults to 0) &#x2014;
The momentum value speeds up the optimizer by taking bigger steps.`,name:"momentum"},{anchor:"bitsandbytes.optim.SGD.__init__.dampening",description:`<strong>dampening</strong> (<code>float</code>, defaults to 0) &#x2014;
The dampening value reduces the momentum of the optimizer.`,name:"dampening"},{anchor:"bitsandbytes.optim.SGD.__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.SGD.__init__.nesterov",description:`<strong>nesterov</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to use Nesterov momentum.`,name:"nesterov"},{anchor:"bitsandbytes.optim.SGD.__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.SGD.__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.SGD.__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"}]}),o(2),n(_),n(s);var g=t(s,2);i(g,{title:"SGD8bit",local:"bitsandbytes.optim.SGD8bit",headingTag:"h2"});var r=t(g,2),h=a(r);e(h,{name:"class bitsandbytes.optim.SGD8bit",anchor:"bitsandbytes.optim.SGD8bit",source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_2038/bitsandbytes/optim/sgd.py#L59",parameters:[{name:"params",val:""},{name:"lr",val:""},{name:"momentum",val:" = 0"},{name:"dampening",val:" = 0"},{name:"weight_decay",val:" = 0"},{name:"nesterov",val:" = False"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"}]});var v=t(h,2),N=a(v);e(N,{name:"__init__",anchor:"bitsandbytes.optim.SGD8bit.__init__",source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_2038/bitsandbytes/optim/sgd.py#L60",parameters:[{name:"params",val:""},{name:"lr",val:""},{name:"momentum",val:" = 0"},{name:"dampening",val:" = 0"},{name:"weight_decay",val:" = 0"},{name:"nesterov",val:" = False"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"}],parametersDescription:[{anchor:"bitsandbytes.optim.SGD8bit.__init__.params",description:`<strong>params</strong> (<code>torch.tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.SGD8bit.__init__.lr",description:`<strong>lr</strong> (<code>float</code>) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.SGD8bit.__init__.momentum",description:`<strong>momentum</strong> (<code>float</code>, defaults to 0) &#x2014;
The momentum value speeds up the optimizer by taking bigger steps.`,name:"momentum"},{anchor:"bitsandbytes.optim.SGD8bit.__init__.dampening",description:`<strong>dampening</strong> (<code>float</code>, defaults to 0) &#x2014;
The dampening value reduces the momentum of the optimizer.`,name:"dampening"},{anchor:"bitsandbytes.optim.SGD8bit.__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.SGD8bit.__init__.nesterov",description:`<strong>nesterov</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to use Nesterov momentum.`,name:"nesterov"},{anchor:"bitsandbytes.optim.SGD8bit.__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.SGD8bit.__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"}]}),o(2),n(v),n(r);var u=t(r,2);i(u,{title:"SGD32bit",local:"bitsandbytes.optim.SGD32bit",headingTag:"h2"});var m=t(u,2),y=a(m);e(y,{name:"class bitsandbytes.optim.SGD32bit",anchor:"bitsandbytes.optim.SGD32bit",source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_2038/bitsandbytes/optim/sgd.py#L107",parameters:[{name:"params",val:""},{name:"lr",val:""},{name:"momentum",val:" = 0"},{name:"dampening",val:" = 0"},{name:"weight_decay",val:" = 0"},{name:"nesterov",val:" = False"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"}]});var f=t(y,2),F=a(f);e(F,{name:"__init__",anchor:"bitsandbytes.optim.SGD32bit.__init__",source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/vr_2038/bitsandbytes/optim/sgd.py#L108",parameters:[{name:"params",val:""},{name:"lr",val:""},{name:"momentum",val:" = 0"},{name:"dampening",val:" = 0"},{name:"weight_decay",val:" = 0"},{name:"nesterov",val:" = False"},{name:"args",val:" = None"},{name:"min_8bit_size",val:" = 4096"}],parametersDescription:[{anchor:"bitsandbytes.optim.SGD32bit.__init__.params",description:`<strong>params</strong> (<code>torch.tensor</code>) &#x2014;
The input parameters to optimize.`,name:"params"},{anchor:"bitsandbytes.optim.SGD32bit.__init__.lr",description:`<strong>lr</strong> (<code>float</code>) &#x2014;
The learning rate.`,name:"lr"},{anchor:"bitsandbytes.optim.SGD32bit.__init__.momentum",description:`<strong>momentum</strong> (<code>float</code>, defaults to 0) &#x2014;
The momentum value speeds up the optimizer by taking bigger steps.`,name:"momentum"},{anchor:"bitsandbytes.optim.SGD32bit.__init__.dampening",description:`<strong>dampening</strong> (<code>float</code>, defaults to 0) &#x2014;
The dampening value reduces the momentum of the optimizer.`,name:"dampening"},{anchor:"bitsandbytes.optim.SGD32bit.__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.SGD32bit.__init__.nesterov",description:`<strong>nesterov</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to use Nesterov momentum.`,name:"nesterov"},{anchor:"bitsandbytes.optim.SGD32bit.__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.SGD32bit.__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"}]}),o(2),n(f),n(m);var L=t(m,2);M(L,{source:"https://github.com/bitsandbytes-foundation/bitsandbytes/blob/main/docs/source/reference/optim/sgd.mdx"}),o(2),G(z,d),E()}export{J as component};