mirrors / facebookresearch--audiocraft

项目文件夹

facebookresearch--audiocraft
文件
gh-docs
T
Alexandre Défossez (autodoc) aeffa9f270 api_docs
2025-03-13 16:07:03 +00:00

882 行
46 KiB
HTML
原始文件 永久链接 Blame 文件历史

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.diffusion_schedule API documentation</title>
<meta name="description" content="Functions for Noise Schedule, defines diffusion process, reverse process and data processor.">
<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.diffusion_schedule</code></h1>
</header>
<section id="section-intro">
<p>Functions for Noise Schedule, defines diffusion process, reverse process and data processor.</p>
</section>
<section>
</section>
<section>
</section>
<section>
<h2 class="section-title" id="header-functions">Functions</h2>
<dl>
<dt id="audiocraft.modules.diffusion_schedule.betas_from_alpha_bar"><code class="name flex">
<span>def <span class="ident">betas_from_alpha_bar</span></span>(<span>alpha_bar)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def betas_from_alpha_bar(alpha_bar):
alphas = torch.cat([torch.Tensor([alpha_bar[0]]), alpha_bar[1:]/alpha_bar[:-1]])
return 1 - alphas</code></pre>
</details>
<div class="desc"></div>
</dd>
</dl>
</section>
<section>
<h2 class="section-title" id="header-classes">Classes</h2>
<dl>
<dt id="audiocraft.modules.diffusion_schedule.MultiBandProcessor"><code class="flex name class">
<span>class <span class="ident">MultiBandProcessor</span></span>
<span>(</span><span>n_bands: int = 8,<br>sample_rate: float = 24000,<br>num_samples: int = 10000,<br>power_std: float | List[float] | torch.Tensor = 1.0)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">class MultiBandProcessor(SampleProcessor):
&#34;&#34;&#34;
MultiBand sample processor. The input audio is splitted across
frequency bands evenly distributed in mel-scale.
Each band will be rescaled to match the power distribution
of Gaussian noise in that band, using online metrics
computed on the first few samples.
Args:
n_bands (int): Number of mel-bands to split the signal over.
sample_rate (int): Sample rate of the audio.
num_samples (int): Number of samples to use to fit the rescaling
for each band. The processor won&#39;t be stable
until it has seen that many samples.
power_std (float or list/tensor): The rescaling factor computed to match the
power of Gaussian noise in each band is taken to
that power, i.e. `1.` means full correction of the energy
in each band, and values less than `1` means only partial
correction. Can be used to balance the relative importance
of low vs. high freq in typical audio signals.
&#34;&#34;&#34;
def __init__(self, n_bands: int = 8, sample_rate: float = 24_000,
num_samples: int = 10_000, power_std: tp.Union[float, tp.List[float], torch.Tensor] = 1.):
super().__init__()
self.n_bands = n_bands
self.split_bands = julius.SplitBands(sample_rate, n_bands=n_bands)
self.num_samples = num_samples
self.power_std = power_std
if isinstance(power_std, list):
assert len(power_std) == n_bands
power_std = torch.tensor(power_std)
self.register_buffer(&#39;counts&#39;, torch.zeros(1))
self.register_buffer(&#39;sum_x&#39;, torch.zeros(n_bands))
self.register_buffer(&#39;sum_x2&#39;, torch.zeros(n_bands))
self.register_buffer(&#39;sum_target_x2&#39;, torch.zeros(n_bands))
self.counts: torch.Tensor
self.sum_x: torch.Tensor
self.sum_x2: torch.Tensor
self.sum_target_x2: torch.Tensor
@property
def mean(self):
mean = self.sum_x / self.counts
return mean
@property
def std(self):
std = (self.sum_x2 / self.counts - self.mean**2).clamp(min=0).sqrt()
return std
@property
def target_std(self):
target_std = self.sum_target_x2 / self.counts
return target_std
def project_sample(self, x: torch.Tensor):
assert x.dim() == 3
bands = self.split_bands(x)
if self.counts.item() &lt; self.num_samples:
ref_bands = self.split_bands(torch.randn_like(x))
self.counts += len(x)
self.sum_x += bands.mean(dim=(2, 3)).sum(dim=1)
self.sum_x2 += bands.pow(2).mean(dim=(2, 3)).sum(dim=1)
self.sum_target_x2 += ref_bands.pow(2).mean(dim=(2, 3)).sum(dim=1)
rescale = (self.target_std / self.std.clamp(min=1e-12)) ** self.power_std # same output size
bands = (bands - self.mean.view(-1, 1, 1, 1)) * rescale.view(-1, 1, 1, 1)
return bands.sum(dim=0)
def return_sample(self, x: torch.Tensor):
assert x.dim() == 3
bands = self.split_bands(x)
rescale = (self.std / self.target_std) ** self.power_std
bands = bands * rescale.view(-1, 1, 1, 1) + self.mean.view(-1, 1, 1, 1)
return bands.sum(dim=0)</code></pre>
</details>
<div class="desc"><p>MultiBand sample processor. The input audio is splitted across
frequency bands evenly distributed in mel-scale.</p>
<p>Each band will be rescaled to match the power distribution
of Gaussian noise in that band, using online metrics
computed on the first few samples.</p>
<h2 id="args">Args</h2>
<dl>
<dt><strong><code>n_bands</code></strong> :&ensp;<code>int</code></dt>
<dd>Number of mel-bands to split the signal over.</dd>
<dt><strong><code>sample_rate</code></strong> :&ensp;<code>int</code></dt>
<dd>Sample rate of the audio.</dd>
<dt><strong><code>num_samples</code></strong> :&ensp;<code>int</code></dt>
<dd>Number of samples to use to fit the rescaling
for each band. The processor won't be stable
until it has seen that many samples.</dd>
</dl>
<p>power_std (float or list/tensor): The rescaling factor computed to match the
power of Gaussian noise in each band is taken to
that power, i.e. <code>1.</code> means full correction of the energy
in each band, and values less than <code>1</code> means only partial
correction. Can be used to balance the relative importance
of low vs. high freq in typical audio signals.
Initializes internal Module state, shared by both nn.Module and ScriptModule.</p></div>
<h3>Ancestors</h3>
<ul class="hlist">
<li><a title="audiocraft.modules.diffusion_schedule.SampleProcessor" href="#audiocraft.modules.diffusion_schedule.SampleProcessor">SampleProcessor</a></li>
<li>torch.nn.modules.module.Module</li>
</ul>
<h3>Class variables</h3>
<dl>
<dt id="audiocraft.modules.diffusion_schedule.MultiBandProcessor.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.diffusion_schedule.MultiBandProcessor.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.diffusion_schedule.MultiBandProcessor.training"><code class="name">var <span class="ident">training</span> : bool</code></dt>
<dd>
<div class="desc"></div>
</dd>
</dl>
<h3>Instance variables</h3>
<dl>
<dt id="audiocraft.modules.diffusion_schedule.MultiBandProcessor.mean"><code class="name">prop <span class="ident">mean</span></code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">@property
def mean(self):
mean = self.sum_x / self.counts
return mean</code></pre>
</details>
<div class="desc"></div>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.MultiBandProcessor.std"><code class="name">prop <span class="ident">std</span></code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">@property
def std(self):
std = (self.sum_x2 / self.counts - self.mean**2).clamp(min=0).sqrt()
return std</code></pre>
</details>
<div class="desc"></div>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.MultiBandProcessor.target_std"><code class="name">prop <span class="ident">target_std</span></code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">@property
def target_std(self):
target_std = self.sum_target_x2 / self.counts
return target_std</code></pre>
</details>
<div class="desc"></div>
</dd>
</dl>
<h3>Inherited members</h3>
<ul class="hlist">
<li><code><b><a title="audiocraft.modules.diffusion_schedule.SampleProcessor" href="#audiocraft.modules.diffusion_schedule.SampleProcessor">SampleProcessor</a></b></code>:
<ul class="hlist">
<li><code><a title="audiocraft.modules.diffusion_schedule.SampleProcessor.forward" href="#audiocraft.modules.diffusion_schedule.SampleProcessor.forward">forward</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.SampleProcessor.project_sample" href="#audiocraft.modules.diffusion_schedule.SampleProcessor.project_sample">project_sample</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.SampleProcessor.return_sample" href="#audiocraft.modules.diffusion_schedule.SampleProcessor.return_sample">return_sample</a></code></li>
</ul>
</li>
</ul>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.NoiseSchedule"><code class="flex name class">
<span>class <span class="ident">NoiseSchedule</span></span>
<span>(</span><span>beta_t0: float = 0.0001,<br>beta_t1: float = 0.02,<br>num_steps: int = 1000,<br>variance: str = 'beta',<br>clip: float = 5.0,<br>rescale: float = 1.0,<br>device='cuda',<br>beta_exp: float = 1,<br>repartition: str = 'power',<br>alpha_sigmoid: dict = {},<br>n_bands: int | None = None,<br>sample_processor: <a title="audiocraft.modules.diffusion_schedule.SampleProcessor" href="#audiocraft.modules.diffusion_schedule.SampleProcessor">SampleProcessor</a> = SampleProcessor(),<br>noise_scale: float = 1.0,<br>**kwargs)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">class NoiseSchedule:
&#34;&#34;&#34;Noise schedule for diffusion.
Args:
beta_t0 (float): Variance of the first diffusion step.
beta_t1 (float): Variance of the last diffusion step.
beta_exp (float): Power schedule exponent
num_steps (int): Number of diffusion step.
variance (str): choice of the sigma value for the denoising eq. Choices: &#34;beta&#34; or &#34;beta_tilde&#34;
clip (float): clipping value for the denoising steps
rescale (float): rescaling value to avoid vanishing signals unused by default (i.e 1)
repartition (str): shape of the schedule only power schedule is supported
sample_processor (SampleProcessor): Module that normalize data to match better the gaussian distribution
noise_scale (float): Scaling factor for the noise
&#34;&#34;&#34;
def __init__(self, beta_t0: float = 1e-4, beta_t1: float = 0.02, num_steps: int = 1000, variance: str = &#39;beta&#39;,
clip: float = 5., rescale: float = 1., device=&#39;cuda&#39;, beta_exp: float = 1,
repartition: str = &#34;power&#34;, alpha_sigmoid: dict = {}, n_bands: tp.Optional[int] = None,
sample_processor: SampleProcessor = SampleProcessor(), noise_scale: float = 1.0, **kwargs):
self.beta_t0 = beta_t0
self.beta_t1 = beta_t1
self.variance = variance
self.num_steps = num_steps
self.clip = clip
self.sample_processor = sample_processor
self.rescale = rescale
self.n_bands = n_bands
self.noise_scale = noise_scale
assert n_bands is None
if repartition == &#34;power&#34;:
self.betas = torch.linspace(beta_t0 ** (1 / beta_exp), beta_t1 ** (1 / beta_exp), num_steps,
device=device, dtype=torch.float) ** beta_exp
else:
raise RuntimeError(&#39;Not implemented&#39;)
self.rng = random.Random(1234)
def get_beta(self, step: tp.Union[int, torch.Tensor]):
if self.n_bands is None:
return self.betas[step]
else:
return self.betas[:, step] # [n_bands, len(step)]
def get_initial_noise(self, x: torch.Tensor):
if self.n_bands is None:
return torch.randn_like(x)
return torch.randn((x.size(0), self.n_bands, x.size(2)))
def get_alpha_bar(self, step: tp.Optional[tp.Union[int, torch.Tensor]] = None) -&gt; torch.Tensor:
&#34;&#34;&#34;Return &#39;alpha_bar&#39;, either for a given step, or as a tensor with its value for each step.&#34;&#34;&#34;
if step is None:
return (1 - self.betas).cumprod(dim=-1) # works for simgle and multi bands
if type(step) is int:
return (1 - self.betas[:step + 1]).prod()
else:
return (1 - self.betas).cumprod(dim=0)[step].view(-1, 1, 1)
def get_training_item(self, x: torch.Tensor, tensor_step: bool = False) -&gt; TrainingItem:
&#34;&#34;&#34;Create a noisy data item for diffusion model training:
Args:
x (torch.Tensor): clean audio data torch.tensor(bs, 1, T)
tensor_step (bool): If tensor_step = false, only one step t is sample,
the whole batch is diffused to the same step and t is int.
If tensor_step = true, t is a tensor of size (x.size(0),)
every element of the batch is diffused to a independently sampled.
&#34;&#34;&#34;
step: tp.Union[int, torch.Tensor]
if tensor_step:
bs = x.size(0)
step = torch.randint(0, self.num_steps, size=(bs,), device=x.device)
else:
step = self.rng.randrange(self.num_steps)
alpha_bar = self.get_alpha_bar(step) # [batch_size, n_bands, 1]
x = self.sample_processor.project_sample(x)
noise = torch.randn_like(x)
noisy = (alpha_bar.sqrt() / self.rescale) * x + (1 - alpha_bar).sqrt() * noise * self.noise_scale
return TrainingItem(noisy, noise, step)
def generate(self, model: torch.nn.Module, initial: tp.Optional[torch.Tensor] = None,
condition: tp.Optional[torch.Tensor] = None, return_list: bool = False):
&#34;&#34;&#34;Full ddpm reverse process.
Args:
model (nn.Module): Diffusion model.
initial (tensor): Initial Noise.
condition (tensor): Input conditionning Tensor (e.g. encodec compressed representation).
return_list (bool): Whether to return the whole process or only the sampled point.
&#34;&#34;&#34;
alpha_bar = self.get_alpha_bar(step=self.num_steps - 1)
current = initial
iterates = [initial]
for step in range(self.num_steps)[::-1]:
with torch.no_grad():
estimate = model(current, step, condition=condition).sample
alpha = 1 - self.betas[step]
previous = (current - (1 - alpha) / (1 - alpha_bar).sqrt() * estimate) / alpha.sqrt()
previous_alpha_bar = self.get_alpha_bar(step=step - 1)
if step == 0:
sigma2 = 0
elif self.variance == &#39;beta&#39;:
sigma2 = 1 - alpha
elif self.variance == &#39;beta_tilde&#39;:
sigma2 = (1 - previous_alpha_bar) / (1 - alpha_bar) * (1 - alpha)
elif self.variance == &#39;none&#39;:
sigma2 = 0
else:
raise ValueError(f&#39;Invalid variance type {self.variance}&#39;)
if sigma2 &gt; 0:
previous += sigma2**0.5 * torch.randn_like(previous) * self.noise_scale
if self.clip:
previous = previous.clamp(-self.clip, self.clip)
current = previous
alpha_bar = previous_alpha_bar
if step == 0:
previous *= self.rescale
if return_list:
iterates.append(previous.cpu())
if return_list:
return iterates
else:
return self.sample_processor.return_sample(previous)
def generate_subsampled(self, model: torch.nn.Module, initial: torch.Tensor, step_list: tp.Optional[list] = None,
condition: tp.Optional[torch.Tensor] = None, return_list: bool = False):
&#34;&#34;&#34;Reverse process that only goes through Markov chain states in step_list.&#34;&#34;&#34;
if step_list is None:
step_list = list(range(1000))[::-50] + [0]
alpha_bar = self.get_alpha_bar(step=self.num_steps - 1)
alpha_bars_subsampled = (1 - self.betas).cumprod(dim=0)[list(reversed(step_list))].cpu()
betas_subsampled = betas_from_alpha_bar(alpha_bars_subsampled)
current = initial * self.noise_scale
iterates = [current]
for idx, step in enumerate(step_list[:-1]):
with torch.no_grad():
estimate = model(current, step, condition=condition).sample * self.noise_scale
alpha = 1 - betas_subsampled[-1 - idx]
previous = (current - (1 - alpha) / (1 - alpha_bar).sqrt() * estimate) / alpha.sqrt()
previous_alpha_bar = self.get_alpha_bar(step_list[idx + 1])
if step == step_list[-2]:
sigma2 = 0
previous_alpha_bar = torch.tensor(1.0)
else:
sigma2 = (1 - previous_alpha_bar) / (1 - alpha_bar) * (1 - alpha)
if sigma2 &gt; 0:
previous += sigma2**0.5 * torch.randn_like(previous) * self.noise_scale
if self.clip:
previous = previous.clamp(-self.clip, self.clip)
current = previous
alpha_bar = previous_alpha_bar
if step == 0:
previous *= self.rescale
if return_list:
iterates.append(previous.cpu())
if return_list:
return iterates
else:
return self.sample_processor.return_sample(previous)</code></pre>
</details>
<div class="desc"><p>Noise schedule for diffusion.</p>
<h2 id="args">Args</h2>
<dl>
<dt><strong><code>beta_t0</code></strong> :&ensp;<code>float</code></dt>
<dd>Variance of the first diffusion step.</dd>
<dt><strong><code>beta_t1</code></strong> :&ensp;<code>float</code></dt>
<dd>Variance of the last diffusion step.</dd>
<dt><strong><code>beta_exp</code></strong> :&ensp;<code>float</code></dt>
<dd>Power schedule exponent</dd>
<dt><strong><code>num_steps</code></strong> :&ensp;<code>int</code></dt>
<dd>Number of diffusion step.</dd>
<dt><strong><code>variance</code></strong> :&ensp;<code>str</code></dt>
<dd>choice of the sigma value for the denoising eq. Choices: "beta" or "beta_tilde"</dd>
<dt><strong><code>clip</code></strong> :&ensp;<code>float</code></dt>
<dd>clipping value for the denoising steps</dd>
<dt><strong><code>rescale</code></strong> :&ensp;<code>float</code></dt>
<dd>rescaling value to avoid vanishing signals unused by default (i.e 1)</dd>
<dt><strong><code>repartition</code></strong> :&ensp;<code>str</code></dt>
<dd>shape of the schedule only power schedule is supported</dd>
<dt><strong><code>sample_processor</code></strong> :&ensp;<code><a title="audiocraft.modules.diffusion_schedule.SampleProcessor" href="#audiocraft.modules.diffusion_schedule.SampleProcessor">SampleProcessor</a></code></dt>
<dd>Module that normalize data to match better the gaussian distribution</dd>
<dt><strong><code>noise_scale</code></strong> :&ensp;<code>float</code></dt>
<dd>Scaling factor for the noise</dd>
</dl></div>
<h3>Methods</h3>
<dl>
<dt id="audiocraft.modules.diffusion_schedule.NoiseSchedule.generate"><code class="name flex">
<span>def <span class="ident">generate</span></span>(<span>self,<br>model: torch.nn.modules.module.Module,<br>initial: torch.Tensor | None = None,<br>condition: torch.Tensor | None = None,<br>return_list: bool = False)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def generate(self, model: torch.nn.Module, initial: tp.Optional[torch.Tensor] = None,
condition: tp.Optional[torch.Tensor] = None, return_list: bool = False):
&#34;&#34;&#34;Full ddpm reverse process.
Args:
model (nn.Module): Diffusion model.
initial (tensor): Initial Noise.
condition (tensor): Input conditionning Tensor (e.g. encodec compressed representation).
return_list (bool): Whether to return the whole process or only the sampled point.
&#34;&#34;&#34;
alpha_bar = self.get_alpha_bar(step=self.num_steps - 1)
current = initial
iterates = [initial]
for step in range(self.num_steps)[::-1]:
with torch.no_grad():
estimate = model(current, step, condition=condition).sample
alpha = 1 - self.betas[step]
previous = (current - (1 - alpha) / (1 - alpha_bar).sqrt() * estimate) / alpha.sqrt()
previous_alpha_bar = self.get_alpha_bar(step=step - 1)
if step == 0:
sigma2 = 0
elif self.variance == &#39;beta&#39;:
sigma2 = 1 - alpha
elif self.variance == &#39;beta_tilde&#39;:
sigma2 = (1 - previous_alpha_bar) / (1 - alpha_bar) * (1 - alpha)
elif self.variance == &#39;none&#39;:
sigma2 = 0
else:
raise ValueError(f&#39;Invalid variance type {self.variance}&#39;)
if sigma2 &gt; 0:
previous += sigma2**0.5 * torch.randn_like(previous) * self.noise_scale
if self.clip:
previous = previous.clamp(-self.clip, self.clip)
current = previous
alpha_bar = previous_alpha_bar
if step == 0:
previous *= self.rescale
if return_list:
iterates.append(previous.cpu())
if return_list:
return iterates
else:
return self.sample_processor.return_sample(previous)</code></pre>
</details>
<div class="desc"><p>Full ddpm reverse process.</p>
<h2 id="args">Args</h2>
<dl>
<dt><strong><code>model</code></strong> :&ensp;<code>nn.Module</code></dt>
<dd>Diffusion model.</dd>
<dt><strong><code>initial</code></strong> :&ensp;<code>tensor</code></dt>
<dd>Initial Noise.</dd>
<dt><strong><code>condition</code></strong> :&ensp;<code>tensor</code></dt>
<dd>Input conditionning Tensor (e.g. encodec compressed representation).</dd>
<dt><strong><code>return_list</code></strong> :&ensp;<code>bool</code></dt>
<dd>Whether to return the whole process or only the sampled point.</dd>
</dl></div>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.NoiseSchedule.generate_subsampled"><code class="name flex">
<span>def <span class="ident">generate_subsampled</span></span>(<span>self,<br>model: torch.nn.modules.module.Module,<br>initial: torch.Tensor,<br>step_list: list | None = None,<br>condition: torch.Tensor | None = None,<br>return_list: bool = False)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def generate_subsampled(self, model: torch.nn.Module, initial: torch.Tensor, step_list: tp.Optional[list] = None,
condition: tp.Optional[torch.Tensor] = None, return_list: bool = False):
&#34;&#34;&#34;Reverse process that only goes through Markov chain states in step_list.&#34;&#34;&#34;
if step_list is None:
step_list = list(range(1000))[::-50] + [0]
alpha_bar = self.get_alpha_bar(step=self.num_steps - 1)
alpha_bars_subsampled = (1 - self.betas).cumprod(dim=0)[list(reversed(step_list))].cpu()
betas_subsampled = betas_from_alpha_bar(alpha_bars_subsampled)
current = initial * self.noise_scale
iterates = [current]
for idx, step in enumerate(step_list[:-1]):
with torch.no_grad():
estimate = model(current, step, condition=condition).sample * self.noise_scale
alpha = 1 - betas_subsampled[-1 - idx]
previous = (current - (1 - alpha) / (1 - alpha_bar).sqrt() * estimate) / alpha.sqrt()
previous_alpha_bar = self.get_alpha_bar(step_list[idx + 1])
if step == step_list[-2]:
sigma2 = 0
previous_alpha_bar = torch.tensor(1.0)
else:
sigma2 = (1 - previous_alpha_bar) / (1 - alpha_bar) * (1 - alpha)
if sigma2 &gt; 0:
previous += sigma2**0.5 * torch.randn_like(previous) * self.noise_scale
if self.clip:
previous = previous.clamp(-self.clip, self.clip)
current = previous
alpha_bar = previous_alpha_bar
if step == 0:
previous *= self.rescale
if return_list:
iterates.append(previous.cpu())
if return_list:
return iterates
else:
return self.sample_processor.return_sample(previous)</code></pre>
</details>
<div class="desc"><p>Reverse process that only goes through Markov chain states in step_list.</p></div>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.NoiseSchedule.get_alpha_bar"><code class="name flex">
<span>def <span class="ident">get_alpha_bar</span></span>(<span>self, step: int | torch.Tensor | None = None) ‑> torch.Tensor</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def get_alpha_bar(self, step: tp.Optional[tp.Union[int, torch.Tensor]] = None) -&gt; torch.Tensor:
&#34;&#34;&#34;Return &#39;alpha_bar&#39;, either for a given step, or as a tensor with its value for each step.&#34;&#34;&#34;
if step is None:
return (1 - self.betas).cumprod(dim=-1) # works for simgle and multi bands
if type(step) is int:
return (1 - self.betas[:step + 1]).prod()
else:
return (1 - self.betas).cumprod(dim=0)[step].view(-1, 1, 1)</code></pre>
</details>
<div class="desc"><p>Return 'alpha_bar', either for a given step, or as a tensor with its value for each step.</p></div>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.NoiseSchedule.get_beta"><code class="name flex">
<span>def <span class="ident">get_beta</span></span>(<span>self, step: int | torch.Tensor)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def get_beta(self, step: tp.Union[int, torch.Tensor]):
if self.n_bands is None:
return self.betas[step]
else:
return self.betas[:, step] # [n_bands, len(step)]</code></pre>
</details>
<div class="desc"></div>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.NoiseSchedule.get_initial_noise"><code class="name flex">
<span>def <span class="ident">get_initial_noise</span></span>(<span>self, x: torch.Tensor)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def get_initial_noise(self, x: torch.Tensor):
if self.n_bands is None:
return torch.randn_like(x)
return torch.randn((x.size(0), self.n_bands, x.size(2)))</code></pre>
</details>
<div class="desc"></div>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.NoiseSchedule.get_training_item"><code class="name flex">
<span>def <span class="ident">get_training_item</span></span>(<span>self, x: torch.Tensor, tensor_step: bool = False) ‑> <a title="audiocraft.modules.diffusion_schedule.TrainingItem" href="#audiocraft.modules.diffusion_schedule.TrainingItem">TrainingItem</a></span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def get_training_item(self, x: torch.Tensor, tensor_step: bool = False) -&gt; TrainingItem:
&#34;&#34;&#34;Create a noisy data item for diffusion model training:
Args:
x (torch.Tensor): clean audio data torch.tensor(bs, 1, T)
tensor_step (bool): If tensor_step = false, only one step t is sample,
the whole batch is diffused to the same step and t is int.
If tensor_step = true, t is a tensor of size (x.size(0),)
every element of the batch is diffused to a independently sampled.
&#34;&#34;&#34;
step: tp.Union[int, torch.Tensor]
if tensor_step:
bs = x.size(0)
step = torch.randint(0, self.num_steps, size=(bs,), device=x.device)
else:
step = self.rng.randrange(self.num_steps)
alpha_bar = self.get_alpha_bar(step) # [batch_size, n_bands, 1]
x = self.sample_processor.project_sample(x)
noise = torch.randn_like(x)
noisy = (alpha_bar.sqrt() / self.rescale) * x + (1 - alpha_bar).sqrt() * noise * self.noise_scale
return TrainingItem(noisy, noise, step)</code></pre>
</details>
<div class="desc"><p>Create a noisy data item for diffusion model training:</p>
<h2 id="args">Args</h2>
<dl>
<dt><strong><code>x</code></strong> :&ensp;<code>torch.Tensor</code></dt>
<dd>clean audio data torch.tensor(bs, 1, T)</dd>
<dt><strong><code>tensor_step</code></strong> :&ensp;<code>bool</code></dt>
<dd>If tensor_step = false, only one step t is sample,
the whole batch is diffused to the same step and t is int.
If tensor_step = true, t is a tensor of size (x.size(0),)
every element of the batch is diffused to a independently sampled.</dd>
</dl></div>
</dd>
</dl>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.SampleProcessor"><code class="flex name class">
<span>class <span class="ident">SampleProcessor</span></span>
<span>(</span><span>*args, **kwargs)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">class SampleProcessor(torch.nn.Module):
def project_sample(self, x: torch.Tensor):
&#34;&#34;&#34;Project the original sample to the &#39;space&#39; where the diffusion will happen.&#34;&#34;&#34;
return x
def return_sample(self, z: torch.Tensor):
&#34;&#34;&#34;Project back from diffusion space to the actual sample space.&#34;&#34;&#34;
return z</code></pre>
</details>
<div class="desc"><p>Base class for all neural network modules.</p>
<p>Your models should also subclass this class.</p>
<p>Modules can also contain other Modules, allowing to nest them in
a tree structure. You can assign the submodules as regular attributes::</p>
<pre><code>import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(1, 20, 5)
self.conv2 = nn.Conv2d(20, 20, 5)
def forward(self, x):
x = F.relu(self.conv1(x))
return F.relu(self.conv2(x))
</code></pre>
<p>Submodules assigned in this way will be registered, and will have their
parameters converted too when you call :meth:<code>to</code>, etc.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>As per the example above, an <code>__init__()</code> call to the parent class
must be made before assignment on the child.</p>
</div>
<p>:ivar training: Boolean represents whether this module is in training or
evaluation mode.
:vartype training: bool</p>
<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>Subclasses</h3>
<ul class="hlist">
<li><a title="audiocraft.modules.diffusion_schedule.MultiBandProcessor" href="#audiocraft.modules.diffusion_schedule.MultiBandProcessor">MultiBandProcessor</a></li>
</ul>
<h3>Class variables</h3>
<dl>
<dt id="audiocraft.modules.diffusion_schedule.SampleProcessor.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.diffusion_schedule.SampleProcessor.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.diffusion_schedule.SampleProcessor.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.diffusion_schedule.SampleProcessor.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.diffusion_schedule.SampleProcessor.project_sample"><code class="name flex">
<span>def <span class="ident">project_sample</span></span>(<span>self, x: torch.Tensor)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def project_sample(self, x: torch.Tensor):
&#34;&#34;&#34;Project the original sample to the &#39;space&#39; where the diffusion will happen.&#34;&#34;&#34;
return x</code></pre>
</details>
<div class="desc"><p>Project the original sample to the 'space' where the diffusion will happen.</p></div>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.SampleProcessor.return_sample"><code class="name flex">
<span>def <span class="ident">return_sample</span></span>(<span>self, z: torch.Tensor)</span>
</code></dt>
<dd>
<details class="source">
<summary>
<span>Expand source code</span>
</summary>
<pre><code class="python">def return_sample(self, z: torch.Tensor):
&#34;&#34;&#34;Project back from diffusion space to the actual sample space.&#34;&#34;&#34;
return z</code></pre>
</details>
<div class="desc"><p>Project back from diffusion space to the actual sample space.</p></div>
</dd>
</dl>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.TrainingItem"><code class="flex name class">
<span>class <span class="ident">TrainingItem</span></span>
<span>(</span><span>noisy, noise, step)</span>
</code></dt>
<dd>
<div class="desc"><p>TrainingItem(noisy, noise, step)</p></div>
<h3>Ancestors</h3>
<ul class="hlist">
<li>builtins.tuple</li>
</ul>
<h3>Instance variables</h3>
<dl>
<dt id="audiocraft.modules.diffusion_schedule.TrainingItem.noise"><code class="name">var <span class="ident">noise</span></code></dt>
<dd>
<div class="desc"><p>Alias for field number 1</p></div>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.TrainingItem.noisy"><code class="name">var <span class="ident">noisy</span></code></dt>
<dd>
<div class="desc"><p>Alias for field number 0</p></div>
</dd>
<dt id="audiocraft.modules.diffusion_schedule.TrainingItem.step"><code class="name">var <span class="ident">step</span></code></dt>
<dd>
<div class="desc"><p>Alias for field number 2</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-functions">Functions</a></h3>
<ul class="">
<li><code><a title="audiocraft.modules.diffusion_schedule.betas_from_alpha_bar" href="#audiocraft.modules.diffusion_schedule.betas_from_alpha_bar">betas_from_alpha_bar</a></code></li>
</ul>
</li>
<li><h3><a href="#header-classes">Classes</a></h3>
<ul>
<li>
<h4><code><a title="audiocraft.modules.diffusion_schedule.MultiBandProcessor" href="#audiocraft.modules.diffusion_schedule.MultiBandProcessor">MultiBandProcessor</a></code></h4>
<ul class="two-column">
<li><code><a title="audiocraft.modules.diffusion_schedule.MultiBandProcessor.call_super_init" href="#audiocraft.modules.diffusion_schedule.MultiBandProcessor.call_super_init">call_super_init</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.MultiBandProcessor.dump_patches" href="#audiocraft.modules.diffusion_schedule.MultiBandProcessor.dump_patches">dump_patches</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.MultiBandProcessor.mean" href="#audiocraft.modules.diffusion_schedule.MultiBandProcessor.mean">mean</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.MultiBandProcessor.std" href="#audiocraft.modules.diffusion_schedule.MultiBandProcessor.std">std</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.MultiBandProcessor.target_std" href="#audiocraft.modules.diffusion_schedule.MultiBandProcessor.target_std">target_std</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.MultiBandProcessor.training" href="#audiocraft.modules.diffusion_schedule.MultiBandProcessor.training">training</a></code></li>
</ul>
</li>
<li>
<h4><code><a title="audiocraft.modules.diffusion_schedule.NoiseSchedule" href="#audiocraft.modules.diffusion_schedule.NoiseSchedule">NoiseSchedule</a></code></h4>
<ul class="two-column">
<li><code><a title="audiocraft.modules.diffusion_schedule.NoiseSchedule.generate" href="#audiocraft.modules.diffusion_schedule.NoiseSchedule.generate">generate</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.NoiseSchedule.generate_subsampled" href="#audiocraft.modules.diffusion_schedule.NoiseSchedule.generate_subsampled">generate_subsampled</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.NoiseSchedule.get_alpha_bar" href="#audiocraft.modules.diffusion_schedule.NoiseSchedule.get_alpha_bar">get_alpha_bar</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.NoiseSchedule.get_beta" href="#audiocraft.modules.diffusion_schedule.NoiseSchedule.get_beta">get_beta</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.NoiseSchedule.get_initial_noise" href="#audiocraft.modules.diffusion_schedule.NoiseSchedule.get_initial_noise">get_initial_noise</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.NoiseSchedule.get_training_item" href="#audiocraft.modules.diffusion_schedule.NoiseSchedule.get_training_item">get_training_item</a></code></li>
</ul>
</li>
<li>
<h4><code><a title="audiocraft.modules.diffusion_schedule.SampleProcessor" href="#audiocraft.modules.diffusion_schedule.SampleProcessor">SampleProcessor</a></code></h4>
<ul class="two-column">
<li><code><a title="audiocraft.modules.diffusion_schedule.SampleProcessor.call_super_init" href="#audiocraft.modules.diffusion_schedule.SampleProcessor.call_super_init">call_super_init</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.SampleProcessor.dump_patches" href="#audiocraft.modules.diffusion_schedule.SampleProcessor.dump_patches">dump_patches</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.SampleProcessor.forward" href="#audiocraft.modules.diffusion_schedule.SampleProcessor.forward">forward</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.SampleProcessor.project_sample" href="#audiocraft.modules.diffusion_schedule.SampleProcessor.project_sample">project_sample</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.SampleProcessor.return_sample" href="#audiocraft.modules.diffusion_schedule.SampleProcessor.return_sample">return_sample</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.SampleProcessor.training" href="#audiocraft.modules.diffusion_schedule.SampleProcessor.training">training</a></code></li>
</ul>
</li>
<li>
<h4><code><a title="audiocraft.modules.diffusion_schedule.TrainingItem" href="#audiocraft.modules.diffusion_schedule.TrainingItem">TrainingItem</a></code></h4>
<ul class="">
<li><code><a title="audiocraft.modules.diffusion_schedule.TrainingItem.noise" href="#audiocraft.modules.diffusion_schedule.TrainingItem.noise">noise</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.TrainingItem.noisy" href="#audiocraft.modules.diffusion_schedule.TrainingItem.noisy">noisy</a></code></li>
<li><code><a title="audiocraft.modules.diffusion_schedule.TrainingItem.step" href="#audiocraft.modules.diffusion_schedule.TrainingItem.step">step</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>
在新工单中引用 查看 Git Blame 复制永久链接