项目文件夹

文件
Alexandre Défossez (autodoc) aeffa9f270 api_docs
2025-03-13 16:07:03 +00:00

484 行
26 KiB
HTML

此文件含有不可见的 Unicode 字符
此文件含有人类无法区分的不可见的 Unicode 字符,但可以由计算机进行不同的处理。 如果您是想特意这样的,可以安全地忽略该警告。 使用 Escape 按钮显示他们。
此文件含有可能会与其他字符混淆的 Unicode 字符。 如果您是想特意这样的,可以安全地忽略该警告。 使用 Escape 按钮显示他们。
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1, minimum-scale=1">
<meta name="generator" content="pdoc3 0.11.5">
<title>audiocraft.modules.rope API documentation</title>
<meta name="description" content="">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/10up-sanitize.css/13.0.0/sanitize.min.css" integrity="sha512-y1dtMcuvtTMJc1yPgEqF0ZjQbhnc/bFhyvIyVNb9Zk5mIGtqVaAB1Ttl28su8AvFMOY0EwRbAe+HCLqj6W7/KA==" crossorigin>
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/10up-sanitize.css/13.0.0/typography.min.css" integrity="sha512-Y1DYSb995BAfxobCkKepB1BqJJTPrOp3zPL74AWFugHHmmdcvO+C48WLrUOlhGMc0QG7AE3f7gmvvcrmX2fDoA==" crossorigin>
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.9.0/styles/default.min.css" crossorigin>
<style>:root{--highlight-color:#fe9}.flex{display:flex !important}body{line-height:1.5em}#content{padding:20px}#sidebar{padding:1.5em;overflow:hidden}#sidebar > *:last-child{margin-bottom:2cm}.http-server-breadcrumbs{font-size:130%;margin:0 0 15px 0}#footer{font-size:.75em;padding:5px 30px;border-top:1px solid #ddd;text-align:right}#footer p{margin:0 0 0 1em;display:inline-block}#footer p:last-child{margin-right:30px}h1,h2,h3,h4,h5{font-weight:300}h1{font-size:2.5em;line-height:1.1em}h2{font-size:1.75em;margin:2em 0 .50em 0}h3{font-size:1.4em;margin:1.6em 0 .7em 0}h4{margin:0;font-size:105%}h1:target,h2:target,h3:target,h4:target,h5:target,h6:target{background:var(--highlight-color);padding:.2em 0}a{color:#058;text-decoration:none;transition:color .2s ease-in-out}a:visited{color:#503}a:hover{color:#b62}.title code{font-weight:bold}h2[id^="header-"]{margin-top:2em}.ident{color:#900;font-weight:bold}pre code{font-size:.8em;line-height:1.4em;padding:1em;display:block}code{background:#f3f3f3;font-family:"DejaVu Sans Mono",monospace;padding:1px 4px;overflow-wrap:break-word}h1 code{background:transparent}pre{border-top:1px solid #ccc;border-bottom:1px solid #ccc;margin:1em 0}#http-server-module-list{display:flex;flex-flow:column}#http-server-module-list div{display:flex}#http-server-module-list dt{min-width:10%}#http-server-module-list p{margin-top:0}.toc ul,#index{list-style-type:none;margin:0;padding:0}#index code{background:transparent}#index h3{border-bottom:1px solid #ddd}#index ul{padding:0}#index h4{margin-top:.6em;font-weight:bold}@media (min-width:200ex){#index .two-column{column-count:2}}@media (min-width:300ex){#index .two-column{column-count:3}}dl{margin-bottom:2em}dl dl:last-child{margin-bottom:4em}dd{margin:0 0 1em 3em}#header-classes + dl > dd{margin-bottom:3em}dd dd{margin-left:2em}dd p{margin:10px 0}.name{background:#eee;font-size:.85em;padding:5px 10px;display:inline-block;min-width:40%}.name:hover{background:#e0e0e0}dt:target .name{background:var(--highlight-color)}.name > span:first-child{white-space:nowrap}.name.class > span:nth-child(2){margin-left:.4em}.inherited{color:#999;border-left:5px solid #eee;padding-left:1em}.inheritance em{font-style:normal;font-weight:bold}.desc h2{font-weight:400;font-size:1.25em}.desc h3{font-size:1em}.desc dt code{background:inherit}.source > summary,.git-link-div{color:#666;text-align:right;font-weight:400;font-size:.8em;text-transform:uppercase}.source summary > *{white-space:nowrap;cursor:pointer}.git-link{color:inherit;margin-left:1em}.source pre{max-height:500px;overflow:auto;margin:0}.source pre code{font-size:12px;overflow:visible;min-width:max-content}.hlist{list-style:none}.hlist li{display:inline}.hlist li:after{content:',\2002'}.hlist li:last-child:after{content:none}.hlist .hlist{display:inline;padding-left:1em}img{max-width:100%}td{padding:0 .5em}.admonition{padding:.1em 1em;margin:1em 0}.admonition-title{font-weight:bold}.admonition.note,.admonition.info,.admonition.important{background:#aef}.admonition.todo,.admonition.versionadded,.admonition.tip,.admonition.hint{background:#dfd}.admonition.warning,.admonition.versionchanged,.admonition.deprecated{background:#fd4}.admonition.error,.admonition.danger,.admonition.caution{background:lightpink}</style>
<style media="screen and (min-width: 700px)">@media screen and (min-width:700px){#sidebar{width:30%;height:100vh;overflow:auto;position:sticky;top:0}#content{width:70%;max-width:100ch;padding:3em 4em;border-left:1px solid #ddd}pre code{font-size:1em}.name{font-size:1em}main{display:flex;flex-direction:row-reverse;justify-content:flex-end}.toc ul ul,#index ul ul{padding-left:1em}.toc > ul > li{margin-top:.5em}}</style>
<style media="print">@media print{#sidebar h1{page-break-before:always}.source{display:none}}@media print{*{background:transparent !important;color:#000 !important;box-shadow:none !important;text-shadow:none !important}a[href]:after{content:" (" attr(href) ")";font-size:90%}a[href][title]:after{content:none}abbr[title]:after{content:" (" attr(title) ")"}.ir a:after,a[href^="javascript:"]:after,a[href^="#"]:after{content:""}pre,blockquote{border:1px solid #999;page-break-inside:avoid}thead{display:table-header-group}tr,img{page-break-inside:avoid}img{max-width:100% !important}@page{margin:0.5cm}p,h2,h3{orphans:3;widows:3}h1,h2,h3,h4,h5,h6{page-break-after:avoid}}</style>
<script defer src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.9.0/highlight.min.js" integrity="sha512-D9gUyxqja7hBtkWpPWGt9wfbfaMGVt9gnyCvYa+jojwwPHLCzUm5i8rpk7vD7wNee9bA35eYIjobYPaQuKS1MQ==" crossorigin></script>
<script>window.addEventListener('DOMContentLoaded', () => {
hljs.configure({languages: ['bash', 'css', 'diff', 'graphql', 'ini', 'javascript', 'json', 'plaintext', 'python', 'python-repl', 'rust', 'shell', 'sql', 'typescript', 'xml', 'yaml']});
hljs.highlightAll();
/* Collapse source docstrings */
setTimeout(() => {
[...document.querySelectorAll('.hljs.language-python > .hljs-string')]
.filter(el => el.innerHTML.length > 200 && ['"""', "'''"].includes(el.innerHTML.substring(0, 3)))
.forEach(el => {
let d = document.createElement('details');
d.classList.add('hljs-string');
d.innerHTML = '<summary>"""</summary>' + el.innerHTML.substring(3);
el.replaceWith(d);
});
}, 100);
})</script>
</head>
<body>
<main>
<article id="content">
<header>
<h1 class="title">Module <code>audiocraft.modules.rope</code></h1>
</header>
<section id="section-intro">
</section>
<section>
</section>
<section>
</section>
<section>
</section>
<section>
<h2 class="section-title" id="header-classes">Classes</h2>
<dl>
<dt id="audiocraft.modules.rope.RotaryEmbedding"><code class="flex name class">
<span>class <span class="ident">RotaryEmbedding</span></span>
<span>(</span><span>dim: int,<br>max_period: float = 10000.0,<br>xpos: bool = False,<br>scale: float = 1.0,<br>device=None,<br>dtype: torch.dtype = torch.float32)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">class RotaryEmbedding(nn.Module):
&#34;&#34;&#34;Rotary positional embedding (RoPE) from [Su et al 2022](https://arxiv.org/abs/2104.09864).
Args:
dim (int): Embedding dimension (twice the number of frequencies).
max_period (float): Maximum period of the rotation frequencies.
xpos (bool): Use xPos, applies an exponential decay to rotation matrix.
scale (float): Scale of positional embedding, set to 0 to deactivate.
device (torch.device, optional): Device on which to initialize the module.
dtype (torch.dtype): dtype to use to generate the embedding.
&#34;&#34;&#34;
def __init__(self, dim: int, max_period: float = 10000.0, xpos: bool = False,
scale: float = 1.0, device=None, dtype: torch.dtype = torch.float32):
super().__init__()
assert dim % 2 == 0
self.scale = scale
assert dtype in [torch.float64, torch.float32]
self.dtype = dtype
adim = torch.arange(0, dim, 2, device=device, dtype=dtype)[: (dim // 2)]
frequencies = 1.0 / (max_period ** (adim / dim))
self.register_buffer(&#34;frequencies&#34;, frequencies)
self.rotation: tp.Optional[torch.Tensor] = None
self.xpos = XPos(dim, device=device, dtype=dtype) if xpos else None
def get_rotation(self, start: int, end: int):
&#34;&#34;&#34;Create complex rotation tensor, cache values for fast computation.&#34;&#34;&#34;
if self.rotation is None or end &gt; self.rotation.shape[0]:
assert isinstance(self.frequencies, torch.Tensor) # Satisfy type checker.
idx = torch.arange(end, device=self.frequencies.device, dtype=self.dtype)
angles = torch.outer(idx, self.frequencies)
self.rotation = torch.polar(torch.ones_like(angles), angles)
return self.rotation[start:end]
def rotate(self, x: torch.Tensor, start: int = 0, time_dim: int = 1, invert_decay: bool = False):
&#34;&#34;&#34;Apply rope rotation to query or key tensor.&#34;&#34;&#34;
T = x.shape[time_dim]
target_shape = [1] * x.dim()
target_shape[time_dim] = T
target_shape[-1] = -1
rotation = self.get_rotation(start, start + T).view(target_shape)
if self.xpos:
decay = self.xpos.get_decay(start, start + T).view(target_shape)
else:
decay = 1.0
if invert_decay:
decay = decay ** -1
x_complex = torch.view_as_complex(x.to(self.dtype).reshape(*x.shape[:-1], -1, 2))
scaled_rotation = (rotation * decay) * self.scale + (1.0 - self.scale)
x_out = torch.view_as_real(x_complex * scaled_rotation).view_as(x)
return x_out.type_as(x)
def rotate_qk(self, query: torch.Tensor, key: torch.Tensor, start: int = 0, time_dim: int = 1):
&#34;&#34;&#34; Apply rope rotation to both query and key tensors.
Supports streaming mode, in which query and key are not expected to have the same shape.
In streaming mode, key will be of length [P + C] with P the cached past timesteps, but
query will be [C] (typically C == 1).
Args:
query (torch.Tensor): Query to rotate.
key (torch.Tensor): Key to rotate.
start (int): Start index of the sequence for time offset.
time_dim (int): which dimension represent the time steps.
&#34;&#34;&#34;
query_timesteps = query.shape[time_dim]
key_timesteps = key.shape[time_dim]
streaming_offset = key_timesteps - query_timesteps
query_out = self.rotate(query, start + streaming_offset, time_dim)
key_out = self.rotate(key, start, time_dim, invert_decay=True)
return query_out, key_out</code></pre>
</details>
<div class="desc"><p>Rotary positional embedding (RoPE) from <a href="https://arxiv.org/abs/2104.09864">Su et al 2022</a>.</p>
<h2 id="args">Args</h2>
<dl>
<dt><strong><code>dim</code></strong> :&ensp;<code>int</code></dt>
<dd>Embedding dimension (twice the number of frequencies).</dd>
<dt><strong><code>max_period</code></strong> :&ensp;<code>float</code></dt>
<dd>Maximum period of the rotation frequencies.</dd>
<dt><strong><code>xpos</code></strong> :&ensp;<code>bool</code></dt>
<dd>Use xPos, applies an exponential decay to rotation matrix.</dd>
<dt><strong><code>scale</code></strong> :&ensp;<code>float</code></dt>
<dd>Scale of positional embedding, set to 0 to deactivate.</dd>
<dt><strong><code>device</code></strong> :&ensp;<code>torch.device</code>, optional</dt>
<dd>Device on which to initialize the module.</dd>
<dt><strong><code>dtype</code></strong> :&ensp;<code>torch.dtype</code></dt>
<dd>dtype to use to generate the embedding.</dd>
</dl>
<p>Initializes internal Module state, shared by both nn.Module and ScriptModule.</p></div>
<h3>Ancestors</h3>
<ul class="hlist">
<li>torch.nn.modules.module.Module</li>
</ul>
<h3>Class variables</h3>
<dl>
<dt id="audiocraft.modules.rope.RotaryEmbedding.call_super_init"><code class="name">var <span class="ident">call_super_init</span> : bool</code></dt>
<dd>
<div class="desc"></div>
</dd>
<dt id="audiocraft.modules.rope.RotaryEmbedding.dump_patches"><code class="name">var <span class="ident">dump_patches</span> : bool</code></dt>
<dd>
<div class="desc"></div>
</dd>
<dt id="audiocraft.modules.rope.RotaryEmbedding.training"><code class="name">var <span class="ident">training</span> : bool</code></dt>
<dd>
<div class="desc"></div>
</dd>
</dl>
<h3>Methods</h3>
<dl>
<dt id="audiocraft.modules.rope.RotaryEmbedding.forward"><code class="name flex">
<span>def <span class="ident">forward</span></span>(<span>self, *input: Any) > None</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def _forward_unimplemented(self, *input: Any) -&gt; None:
r&#34;&#34;&#34;Defines the computation performed at every call.
Should be overridden by all subclasses.
.. note::
Although the recipe for forward pass needs to be defined within
this function, one should call the :class:`Module` instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.
&#34;&#34;&#34;
raise NotImplementedError(f&#34;Module [{type(self).__name__}] is missing the required \&#34;forward\&#34; function&#34;)</code></pre>
</details>
<div class="desc"><p>Defines the computation performed at every call.</p>
<p>Should be overridden by all subclasses.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Although the recipe for forward pass needs to be defined within
this function, one should call the :class:<code>Module</code> instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.</p>
</div></div>
</dd>
<dt id="audiocraft.modules.rope.RotaryEmbedding.get_rotation"><code class="name flex">
<span>def <span class="ident">get_rotation</span></span>(<span>self, start: int, end: int)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def get_rotation(self, start: int, end: int):
&#34;&#34;&#34;Create complex rotation tensor, cache values for fast computation.&#34;&#34;&#34;
if self.rotation is None or end &gt; self.rotation.shape[0]:
assert isinstance(self.frequencies, torch.Tensor) # Satisfy type checker.
idx = torch.arange(end, device=self.frequencies.device, dtype=self.dtype)
angles = torch.outer(idx, self.frequencies)
self.rotation = torch.polar(torch.ones_like(angles), angles)
return self.rotation[start:end]</code></pre>
</details>
<div class="desc"><p>Create complex rotation tensor, cache values for fast computation.</p></div>
</dd>
<dt id="audiocraft.modules.rope.RotaryEmbedding.rotate"><code class="name flex">
<span>def <span class="ident">rotate</span></span>(<span>self,<br>x: torch.Tensor,<br>start: int = 0,<br>time_dim: int = 1,<br>invert_decay: bool = False)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def rotate(self, x: torch.Tensor, start: int = 0, time_dim: int = 1, invert_decay: bool = False):
&#34;&#34;&#34;Apply rope rotation to query or key tensor.&#34;&#34;&#34;
T = x.shape[time_dim]
target_shape = [1] * x.dim()
target_shape[time_dim] = T
target_shape[-1] = -1
rotation = self.get_rotation(start, start + T).view(target_shape)
if self.xpos:
decay = self.xpos.get_decay(start, start + T).view(target_shape)
else:
decay = 1.0
if invert_decay:
decay = decay ** -1
x_complex = torch.view_as_complex(x.to(self.dtype).reshape(*x.shape[:-1], -1, 2))
scaled_rotation = (rotation * decay) * self.scale + (1.0 - self.scale)
x_out = torch.view_as_real(x_complex * scaled_rotation).view_as(x)
return x_out.type_as(x)</code></pre>
</details>
<div class="desc"><p>Apply rope rotation to query or key tensor.</p></div>
</dd>
<dt id="audiocraft.modules.rope.RotaryEmbedding.rotate_qk"><code class="name flex">
<span>def <span class="ident">rotate_qk</span></span>(<span>self, query: torch.Tensor, key: torch.Tensor, start: int = 0, time_dim: int = 1)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def rotate_qk(self, query: torch.Tensor, key: torch.Tensor, start: int = 0, time_dim: int = 1):
&#34;&#34;&#34; Apply rope rotation to both query and key tensors.
Supports streaming mode, in which query and key are not expected to have the same shape.
In streaming mode, key will be of length [P + C] with P the cached past timesteps, but
query will be [C] (typically C == 1).
Args:
query (torch.Tensor): Query to rotate.
key (torch.Tensor): Key to rotate.
start (int): Start index of the sequence for time offset.
time_dim (int): which dimension represent the time steps.
&#34;&#34;&#34;
query_timesteps = query.shape[time_dim]
key_timesteps = key.shape[time_dim]
streaming_offset = key_timesteps - query_timesteps
query_out = self.rotate(query, start + streaming_offset, time_dim)
key_out = self.rotate(key, start, time_dim, invert_decay=True)
return query_out, key_out</code></pre>
</details>
<div class="desc"><p>Apply rope rotation to both query and key tensors.
Supports streaming mode, in which query and key are not expected to have the same shape.
In streaming mode, key will be of length [P + C] with P the cached past timesteps, but
query will be [C] (typically C == 1).</p>
<h2 id="args">Args</h2>
<dl>
<dt><strong><code>query</code></strong> :&ensp;<code>torch.Tensor</code></dt>
<dd>Query to rotate.</dd>
<dt><strong><code>key</code></strong> :&ensp;<code>torch.Tensor</code></dt>
<dd>Key to rotate.</dd>
<dt><strong><code>start</code></strong> :&ensp;<code>int</code></dt>
<dd>Start index of the sequence for time offset.</dd>
<dt><strong><code>time_dim</code></strong> :&ensp;<code>int</code></dt>
<dd>which dimension represent the time steps.</dd>
</dl></div>
</dd>
</dl>
</dd>
<dt id="audiocraft.modules.rope.XPos"><code class="flex name class">
<span>class <span class="ident">XPos</span></span>
<span>(</span><span>dim: int,<br>smoothing: float = 0.4,<br>base_scale: int = 512,<br>device=None,<br>dtype: torch.dtype = torch.float32)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">class XPos(nn.Module):
&#34;&#34;&#34;Length-extrapolatable positional embedding (xPos) from [Sun et al 2022](https://arxiv.org/abs/2212.10554v1).
This applies an exponential decay to the RoPE rotation matrix.
Args:
dim (int): Embedding dimension.
smoothing (float): Smoothing factor applied to the decay rates.
base_scale (int): Base decay rate, given in terms of scaling time.
device (torch.device, optional): Device on which to initialize the module.
dtype (torch.dtype): dtype to use to generate the embedding.
&#34;&#34;&#34;
def __init__(self, dim: int, smoothing: float = 0.4, base_scale: int = 512,
device=None, dtype: torch.dtype = torch.float32):
super().__init__()
assert dim % 2 == 0
assert dtype in [torch.float64, torch.float32]
self.dtype = dtype
self.base_scale = base_scale
half_dim = dim // 2
adim = torch.arange(half_dim, device=device, dtype=dtype)
decay_rates = (adim / half_dim + smoothing) / (1.0 + smoothing)
self.register_buffer(&#34;decay_rates&#34;, decay_rates)
self.decay: tp.Optional[torch.Tensor] = None
def get_decay(self, start: int, end: int):
&#34;&#34;&#34;Create complex decay tensor, cache values for fast computation.&#34;&#34;&#34;
if self.decay is None or end &gt; self.decay.shape[0]:
assert isinstance(self.decay_rates, torch.Tensor) # Satisfy type checker.
idx = torch.arange(end, device=self.decay_rates.device, dtype=self.dtype)
power = idx / self.base_scale
scale = self.decay_rates ** power.unsqueeze(-1)
self.decay = torch.polar(scale, torch.zeros_like(scale))
return self.decay[start:end] # [T, C/2]</code></pre>
</details>
<div class="desc"><p>Length-extrapolatable positional embedding (xPos) from <a href="https://arxiv.org/abs/2212.10554v1">Sun et al 2022</a>.
This applies an exponential decay to the RoPE rotation matrix.</p>
<h2 id="args">Args</h2>
<dl>
<dt><strong><code>dim</code></strong> :&ensp;<code>int</code></dt>
<dd>Embedding dimension.</dd>
<dt><strong><code>smoothing</code></strong> :&ensp;<code>float</code></dt>
<dd>Smoothing factor applied to the decay rates.</dd>
<dt><strong><code>base_scale</code></strong> :&ensp;<code>int</code></dt>
<dd>Base decay rate, given in terms of scaling time.</dd>
<dt><strong><code>device</code></strong> :&ensp;<code>torch.device</code>, optional</dt>
<dd>Device on which to initialize the module.</dd>
<dt><strong><code>dtype</code></strong> :&ensp;<code>torch.dtype</code></dt>
<dd>dtype to use to generate the embedding.</dd>
</dl>
<p>Initializes internal Module state, shared by both nn.Module and ScriptModule.</p></div>
<h3>Ancestors</h3>
<ul class="hlist">
<li>torch.nn.modules.module.Module</li>
</ul>
<h3>Class variables</h3>
<dl>
<dt id="audiocraft.modules.rope.XPos.call_super_init"><code class="name">var <span class="ident">call_super_init</span> : bool</code></dt>
<dd>
<div class="desc"></div>
</dd>
<dt id="audiocraft.modules.rope.XPos.dump_patches"><code class="name">var <span class="ident">dump_patches</span> : bool</code></dt>
<dd>
<div class="desc"></div>
</dd>
<dt id="audiocraft.modules.rope.XPos.training"><code class="name">var <span class="ident">training</span> : bool</code></dt>
<dd>
<div class="desc"></div>
</dd>
</dl>
<h3>Methods</h3>
<dl>
<dt id="audiocraft.modules.rope.XPos.forward"><code class="name flex">
<span>def <span class="ident">forward</span></span>(<span>self, *input: Any) > None</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def _forward_unimplemented(self, *input: Any) -&gt; None:
r&#34;&#34;&#34;Defines the computation performed at every call.
Should be overridden by all subclasses.
.. note::
Although the recipe for forward pass needs to be defined within
this function, one should call the :class:`Module` instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.
&#34;&#34;&#34;
raise NotImplementedError(f&#34;Module [{type(self).__name__}] is missing the required \&#34;forward\&#34; function&#34;)</code></pre>
</details>
<div class="desc"><p>Defines the computation performed at every call.</p>
<p>Should be overridden by all subclasses.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Although the recipe for forward pass needs to be defined within
this function, one should call the :class:<code>Module</code> instance afterwards
instead of this since the former takes care of running the
registered hooks while the latter silently ignores them.</p>
</div></div>
</dd>
<dt id="audiocraft.modules.rope.XPos.get_decay"><code class="name flex">
<span>def <span class="ident">get_decay</span></span>(<span>self, start: int, end: int)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def get_decay(self, start: int, end: int):
&#34;&#34;&#34;Create complex decay tensor, cache values for fast computation.&#34;&#34;&#34;
if self.decay is None or end &gt; self.decay.shape[0]:
assert isinstance(self.decay_rates, torch.Tensor) # Satisfy type checker.
idx = torch.arange(end, device=self.decay_rates.device, dtype=self.dtype)
power = idx / self.base_scale
scale = self.decay_rates ** power.unsqueeze(-1)
self.decay = torch.polar(scale, torch.zeros_like(scale))
return self.decay[start:end] # [T, C/2]</code></pre>
</details>
<div class="desc"><p>Create complex decay tensor, cache values for fast computation.</p></div>
</dd>
</dl>
</dd>
</dl>
</section>
</article>
<nav id="sidebar">
<div class="toc">
<ul></ul>
</div>
<ul id="index">
<li><h3>Super-module</h3>
<ul>
<li><code><a title="audiocraft.modules" href="index.html">audiocraft.modules</a></code></li>
</ul>
</li>
<li><h3><a href="#header-classes">Classes</a></h3>
<ul>
<li>
<h4><code><a title="audiocraft.modules.rope.RotaryEmbedding" href="#audiocraft.modules.rope.RotaryEmbedding">RotaryEmbedding</a></code></h4>
<ul class="two-column">
<li><code><a title="audiocraft.modules.rope.RotaryEmbedding.call_super_init" href="#audiocraft.modules.rope.RotaryEmbedding.call_super_init">call_super_init</a></code></li>
<li><code><a title="audiocraft.modules.rope.RotaryEmbedding.dump_patches" href="#audiocraft.modules.rope.RotaryEmbedding.dump_patches">dump_patches</a></code></li>
<li><code><a title="audiocraft.modules.rope.RotaryEmbedding.forward" href="#audiocraft.modules.rope.RotaryEmbedding.forward">forward</a></code></li>
<li><code><a title="audiocraft.modules.rope.RotaryEmbedding.get_rotation" href="#audiocraft.modules.rope.RotaryEmbedding.get_rotation">get_rotation</a></code></li>
<li><code><a title="audiocraft.modules.rope.RotaryEmbedding.rotate" href="#audiocraft.modules.rope.RotaryEmbedding.rotate">rotate</a></code></li>
<li><code><a title="audiocraft.modules.rope.RotaryEmbedding.rotate_qk" href="#audiocraft.modules.rope.RotaryEmbedding.rotate_qk">rotate_qk</a></code></li>
<li><code><a title="audiocraft.modules.rope.RotaryEmbedding.training" href="#audiocraft.modules.rope.RotaryEmbedding.training">training</a></code></li>
</ul>
</li>
<li>
<h4><code><a title="audiocraft.modules.rope.XPos" href="#audiocraft.modules.rope.XPos">XPos</a></code></h4>
<ul class="">
<li><code><a title="audiocraft.modules.rope.XPos.call_super_init" href="#audiocraft.modules.rope.XPos.call_super_init">call_super_init</a></code></li>
<li><code><a title="audiocraft.modules.rope.XPos.dump_patches" href="#audiocraft.modules.rope.XPos.dump_patches">dump_patches</a></code></li>
<li><code><a title="audiocraft.modules.rope.XPos.forward" href="#audiocraft.modules.rope.XPos.forward">forward</a></code></li>
<li><code><a title="audiocraft.modules.rope.XPos.get_decay" href="#audiocraft.modules.rope.XPos.get_decay">get_decay</a></code></li>
<li><code><a title="audiocraft.modules.rope.XPos.training" href="#audiocraft.modules.rope.XPos.training">training</a></code></li>
</ul>
</li>
</ul>
</li>
</ul>
</nav>
</main>
<footer id="footer">
<p>Generated by <a href="https://pdoc3.github.io/pdoc" title="pdoc: Python API documentation generator"><cite>pdoc</cite> 0.11.5</a>.</p>
</footer>
</body>
</html>