facebookresearch--audiocraft
Alexandre Défossez (autodoc)
617 行
33 KiB
HTML
617 行
33 KiB
HTML
<!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.solvers.diffusion 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.solvers.diffusion</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.solvers.diffusion.DataProcess"><code class="flex name class">
|
||
<span>class <span class="ident">DataProcess</span></span>
|
||
<span>(</span><span>initial_sr: int = 24000,<br>target_sr: int = 16000,<br>use_resampling: bool = False,<br>use_filter: bool = False,<br>n_bands: int = 4,<br>idx_band: int = 0,<br>device: torch.device = device(type='cpu'),<br>cutoffs=None,<br>boost=False)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">class DataProcess:
|
||
"""Apply filtering or resampling.
|
||
|
||
Args:
|
||
initial_sr (int): Initial sample rate.
|
||
target_sr (int): Target sample rate.
|
||
use_resampling: Whether to use resampling or not.
|
||
use_filter (bool):
|
||
n_bands (int): Number of bands to consider.
|
||
idx_band (int):
|
||
device (torch.device or str):
|
||
cutoffs ():
|
||
boost (bool):
|
||
"""
|
||
def __init__(self, initial_sr: int = 24000, target_sr: int = 16000, use_resampling: bool = False,
|
||
use_filter: bool = False, n_bands: int = 4,
|
||
idx_band: int = 0, device: torch.device = torch.device('cpu'), cutoffs=None, boost=False):
|
||
"""Apply filtering or resampling
|
||
Args:
|
||
initial_sr (int): sample rate of the dataset
|
||
target_sr (int): sample rate after resampling
|
||
use_resampling (bool): whether or not performs resampling
|
||
use_filter (bool): when True filter the data to keep only one frequency band
|
||
n_bands (int): Number of bands used
|
||
cuts (none or list): The cutoff frequencies of the band filtering
|
||
if None then we use mel scale bands.
|
||
idx_band (int): index of the frequency band. 0 are lows ... (n_bands - 1) highs
|
||
boost (bool): make the data scale match our music dataset.
|
||
"""
|
||
assert idx_band < n_bands
|
||
self.idx_band = idx_band
|
||
if use_filter:
|
||
if cutoffs is not None:
|
||
self.filter = julius.SplitBands(sample_rate=initial_sr, cutoffs=cutoffs).to(device)
|
||
else:
|
||
self.filter = julius.SplitBands(sample_rate=initial_sr, n_bands=n_bands).to(device)
|
||
self.use_filter = use_filter
|
||
self.use_resampling = use_resampling
|
||
self.target_sr = target_sr
|
||
self.initial_sr = initial_sr
|
||
self.boost = boost
|
||
|
||
def process_data(self, x, metric=False):
|
||
if x is None:
|
||
return None
|
||
if self.boost:
|
||
x /= torch.clamp(x.std(dim=(1, 2), keepdim=True), min=1e-4)
|
||
x * 0.22
|
||
if self.use_filter and not metric:
|
||
x = self.filter(x)[self.idx_band]
|
||
if self.use_resampling:
|
||
x = julius.resample_frac(x, old_sr=self.initial_sr, new_sr=self.target_sr)
|
||
return x
|
||
|
||
def inverse_process(self, x):
|
||
"""Upsampling only."""
|
||
if self.use_resampling:
|
||
x = julius.resample_frac(x, old_sr=self.target_sr, new_sr=self.target_sr)
|
||
return x</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Apply filtering or resampling.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>initial_sr</code></strong> : <code>int</code></dt>
|
||
<dd>Initial sample rate.</dd>
|
||
<dt><strong><code>target_sr</code></strong> : <code>int</code></dt>
|
||
<dd>Target sample rate.</dd>
|
||
<dt><strong><code>use_resampling</code></strong></dt>
|
||
<dd>Whether to use resampling or not.</dd>
|
||
<dt>use_filter (bool):</dt>
|
||
<dt><strong><code>n_bands</code></strong> : <code>int</code></dt>
|
||
<dd>Number of bands to consider.</dd>
|
||
</dl>
|
||
<p>idx_band (int):
|
||
device (torch.device or str):
|
||
cutoffs ():
|
||
boost (bool):
|
||
Apply filtering or resampling</p>
|
||
<h2 id="args_1">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>initial_sr</code></strong> : <code>int</code></dt>
|
||
<dd>sample rate of the dataset</dd>
|
||
<dt><strong><code>target_sr</code></strong> : <code>int</code></dt>
|
||
<dd>sample rate after resampling</dd>
|
||
<dt><strong><code>use_resampling</code></strong> : <code>bool</code></dt>
|
||
<dd>whether or not performs resampling</dd>
|
||
<dt><strong><code>use_filter</code></strong> : <code>bool</code></dt>
|
||
<dd>when True filter the data to keep only one frequency band</dd>
|
||
<dt><strong><code>n_bands</code></strong> : <code>int</code></dt>
|
||
<dd>Number of bands used</dd>
|
||
<dt><strong><code>cuts</code></strong> : <code>none</code> or <code>list</code></dt>
|
||
<dd>The cutoff frequencies of the band filtering
|
||
if None then we use mel scale bands.</dd>
|
||
<dt><strong><code>idx_band</code></strong> : <code>int</code></dt>
|
||
<dd>index of the frequency band. 0 are lows … (n_bands - 1) highs</dd>
|
||
<dt><strong><code>boost</code></strong> : <code>bool</code></dt>
|
||
<dd>make the data scale match our music dataset.</dd>
|
||
</dl></div>
|
||
<h3>Methods</h3>
|
||
<dl>
|
||
<dt id="audiocraft.solvers.diffusion.DataProcess.inverse_process"><code class="name flex">
|
||
<span>def <span class="ident">inverse_process</span></span>(<span>self, x)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def inverse_process(self, x):
|
||
"""Upsampling only."""
|
||
if self.use_resampling:
|
||
x = julius.resample_frac(x, old_sr=self.target_sr, new_sr=self.target_sr)
|
||
return x</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Upsampling only.</p></div>
|
||
</dd>
|
||
<dt id="audiocraft.solvers.diffusion.DataProcess.process_data"><code class="name flex">
|
||
<span>def <span class="ident">process_data</span></span>(<span>self, x, metric=False)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def process_data(self, x, metric=False):
|
||
if x is None:
|
||
return None
|
||
if self.boost:
|
||
x /= torch.clamp(x.std(dim=(1, 2), keepdim=True), min=1e-4)
|
||
x * 0.22
|
||
if self.use_filter and not metric:
|
||
x = self.filter(x)[self.idx_band]
|
||
if self.use_resampling:
|
||
x = julius.resample_frac(x, old_sr=self.initial_sr, new_sr=self.target_sr)
|
||
return x</code></pre>
|
||
</details>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
</dl>
|
||
</dd>
|
||
<dt id="audiocraft.solvers.diffusion.DiffusionSolver"><code class="flex name class">
|
||
<span>class <span class="ident">DiffusionSolver</span></span>
|
||
<span>(</span><span>cfg: omegaconf.dictconfig.DictConfig)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">class DiffusionSolver(base.StandardSolver):
|
||
"""Solver for compression task.
|
||
|
||
The diffusion task allows for MultiBand diffusion model training.
|
||
|
||
Args:
|
||
cfg (DictConfig): Configuration.
|
||
"""
|
||
def __init__(self, cfg: omegaconf.DictConfig):
|
||
super().__init__(cfg)
|
||
self.cfg = cfg
|
||
self.device = cfg.device
|
||
self.sample_rate: int = self.cfg.sample_rate
|
||
self.codec_model = CompressionSolver.model_from_checkpoint(
|
||
cfg.compression_model_checkpoint, device=self.device)
|
||
|
||
self.codec_model.set_num_codebooks(cfg.n_q)
|
||
assert self.codec_model.sample_rate == self.cfg.sample_rate, (
|
||
f"Codec model sample rate is {self.codec_model.sample_rate} but "
|
||
f"Solver sample rate is {self.cfg.sample_rate}."
|
||
)
|
||
assert self.codec_model.sample_rate == self.sample_rate, \
|
||
f"Sample rate of solver {self.sample_rate} and codec {self.codec_model.sample_rate} " \
|
||
"don't match."
|
||
|
||
self.sample_processor = get_processor(cfg.processor, sample_rate=self.sample_rate)
|
||
self.register_stateful('sample_processor')
|
||
self.sample_processor.to(self.device)
|
||
|
||
self.schedule = NoiseSchedule(
|
||
**cfg.schedule, device=self.device, sample_processor=self.sample_processor)
|
||
|
||
self.eval_metric: tp.Optional[torch.nn.Module] = None
|
||
|
||
self.rvm = RelativeVolumeMel()
|
||
self.data_processor = DataProcess(initial_sr=self.sample_rate, target_sr=cfg.resampling.target_sr,
|
||
use_resampling=cfg.resampling.use, cutoffs=cfg.filter.cutoffs,
|
||
use_filter=cfg.filter.use, n_bands=cfg.filter.n_bands,
|
||
idx_band=cfg.filter.idx_band, device=self.device)
|
||
|
||
@property
|
||
def best_metric_name(self) -> tp.Optional[str]:
|
||
if self._current_stage == "evaluate":
|
||
return 'rvm'
|
||
else:
|
||
return 'loss'
|
||
|
||
@torch.no_grad()
|
||
def get_condition(self, wav: torch.Tensor) -> torch.Tensor:
|
||
codes, scale = self.codec_model.encode(wav)
|
||
assert scale is None, "Scaled compression models not supported."
|
||
emb = self.codec_model.decode_latent(codes)
|
||
return emb
|
||
|
||
def build_model(self):
|
||
"""Build model and optimizer as well as optional Exponential Moving Average of the model.
|
||
"""
|
||
# Model and optimizer
|
||
self.model = models.builders.get_diffusion_model(self.cfg).to(self.device)
|
||
self.optimizer = builders.get_optimizer(self.model.parameters(), self.cfg.optim)
|
||
self.register_stateful('model', 'optimizer')
|
||
self.register_best_state('model')
|
||
self.register_ema('model')
|
||
|
||
def build_dataloaders(self):
|
||
"""Build audio dataloaders for each stage."""
|
||
self.dataloaders = builders.get_audio_datasets(self.cfg)
|
||
|
||
def show(self):
|
||
# TODO
|
||
raise NotImplementedError()
|
||
|
||
def run_step(self, idx: int, batch: torch.Tensor, metrics: dict):
|
||
"""Perform one training or valid step on a given batch."""
|
||
x = batch.to(self.device)
|
||
loss_fun = F.mse_loss if self.cfg.loss.kind == 'mse' else F.l1_loss
|
||
|
||
condition = self.get_condition(x) # [bs, 128, T/hop, n_emb]
|
||
sample = self.data_processor.process_data(x)
|
||
|
||
input_, target, step = self.schedule.get_training_item(sample,
|
||
tensor_step=self.cfg.schedule.variable_step_batch)
|
||
out = self.model(input_, step, condition=condition).sample
|
||
|
||
base_loss = loss_fun(out, target, reduction='none').mean(dim=(1, 2))
|
||
reference_loss = loss_fun(input_, target, reduction='none').mean(dim=(1, 2))
|
||
loss = base_loss / reference_loss ** self.cfg.loss.norm_power
|
||
|
||
if self.is_training:
|
||
loss.mean().backward()
|
||
flashy.distrib.sync_model(self.model)
|
||
self.optimizer.step()
|
||
self.optimizer.zero_grad()
|
||
metrics = {
|
||
'loss': loss.mean(), 'normed_loss': (base_loss / reference_loss).mean(),
|
||
}
|
||
metrics.update(self.per_stage({'loss': loss, 'normed_loss': base_loss / reference_loss}, step))
|
||
metrics.update({
|
||
'std_in': input_.std(), 'std_out': out.std()})
|
||
return metrics
|
||
|
||
def run_epoch(self):
|
||
# reset random seed at the beginning of the epoch
|
||
self.rng = torch.Generator()
|
||
self.rng.manual_seed(1234 + self.epoch)
|
||
self.per_stage = PerStageMetrics(self.schedule.num_steps, self.cfg.metrics.num_stage)
|
||
# run epoch
|
||
super().run_epoch()
|
||
|
||
def evaluate(self):
|
||
"""Evaluate stage.
|
||
Runs audio reconstruction evaluation.
|
||
"""
|
||
self.model.eval()
|
||
evaluate_stage_name = f'{self.current_stage}'
|
||
loader = self.dataloaders['evaluate']
|
||
updates = len(loader)
|
||
lp = self.log_progress(f'{evaluate_stage_name} estimate', loader, total=updates, updates=self.log_updates)
|
||
|
||
metrics = {}
|
||
n = 1
|
||
for idx, batch in enumerate(lp):
|
||
x = batch.to(self.device)
|
||
with torch.no_grad():
|
||
y_pred = self.regenerate(x)
|
||
|
||
y_pred = y_pred.cpu()
|
||
y = batch.cpu() # should already be on CPU but just in case
|
||
rvm = self.rvm(y_pred, y)
|
||
lp.update(**rvm)
|
||
if len(metrics) == 0:
|
||
metrics = rvm
|
||
else:
|
||
for key in rvm.keys():
|
||
metrics[key] = (metrics[key] * n + rvm[key]) / (n + 1)
|
||
metrics = flashy.distrib.average_metrics(metrics)
|
||
return metrics
|
||
|
||
@torch.no_grad()
|
||
def regenerate(self, wav: torch.Tensor, step_list: tp.Optional[list] = None):
|
||
"""Regenerate the given waveform."""
|
||
condition = self.get_condition(wav)
|
||
initial = self.schedule.get_initial_noise(self.data_processor.process_data(wav)) # sampling rate changes.
|
||
result = self.schedule.generate_subsampled(self.model, initial=initial, condition=condition,
|
||
step_list=step_list)
|
||
result = self.data_processor.inverse_process(result)
|
||
return result
|
||
|
||
def generate(self):
|
||
"""Generate stage."""
|
||
sample_manager = SampleManager(self.xp)
|
||
self.model.eval()
|
||
generate_stage_name = f'{self.current_stage}'
|
||
|
||
loader = self.dataloaders['generate']
|
||
updates = len(loader)
|
||
lp = self.log_progress(generate_stage_name, loader, total=updates, updates=self.log_updates)
|
||
|
||
for batch in lp:
|
||
reference, _ = batch
|
||
reference = reference.to(self.device)
|
||
estimate = self.regenerate(reference)
|
||
reference = reference.cpu()
|
||
estimate = estimate.cpu()
|
||
sample_manager.add_samples(estimate, self.epoch, ground_truth_wavs=reference)
|
||
flashy.distrib.barrier()</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Solver for compression task.</p>
|
||
<p>The diffusion task allows for MultiBand diffusion model training.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>cfg</code></strong> : <code>DictConfig</code></dt>
|
||
<dd>Configuration.</dd>
|
||
</dl></div>
|
||
<h3>Ancestors</h3>
|
||
<ul class="hlist">
|
||
<li><a title="audiocraft.solvers.base.StandardSolver" href="base.html#audiocraft.solvers.base.StandardSolver">StandardSolver</a></li>
|
||
<li>abc.ABC</li>
|
||
<li>flashy.solver.BaseSolver</li>
|
||
</ul>
|
||
<h3>Methods</h3>
|
||
<dl>
|
||
<dt id="audiocraft.solvers.diffusion.DiffusionSolver.build_dataloaders"><code class="name flex">
|
||
<span>def <span class="ident">build_dataloaders</span></span>(<span>self)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def build_dataloaders(self):
|
||
"""Build audio dataloaders for each stage."""
|
||
self.dataloaders = builders.get_audio_datasets(self.cfg)</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Build audio dataloaders for each stage.</p></div>
|
||
</dd>
|
||
<dt id="audiocraft.solvers.diffusion.DiffusionSolver.build_model"><code class="name flex">
|
||
<span>def <span class="ident">build_model</span></span>(<span>self)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def build_model(self):
|
||
"""Build model and optimizer as well as optional Exponential Moving Average of the model.
|
||
"""
|
||
# Model and optimizer
|
||
self.model = models.builders.get_diffusion_model(self.cfg).to(self.device)
|
||
self.optimizer = builders.get_optimizer(self.model.parameters(), self.cfg.optim)
|
||
self.register_stateful('model', 'optimizer')
|
||
self.register_best_state('model')
|
||
self.register_ema('model')</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Build model and optimizer as well as optional Exponential Moving Average of the model.</p></div>
|
||
</dd>
|
||
<dt id="audiocraft.solvers.diffusion.DiffusionSolver.evaluate"><code class="name flex">
|
||
<span>def <span class="ident">evaluate</span></span>(<span>self)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def evaluate(self):
|
||
"""Evaluate stage.
|
||
Runs audio reconstruction evaluation.
|
||
"""
|
||
self.model.eval()
|
||
evaluate_stage_name = f'{self.current_stage}'
|
||
loader = self.dataloaders['evaluate']
|
||
updates = len(loader)
|
||
lp = self.log_progress(f'{evaluate_stage_name} estimate', loader, total=updates, updates=self.log_updates)
|
||
|
||
metrics = {}
|
||
n = 1
|
||
for idx, batch in enumerate(lp):
|
||
x = batch.to(self.device)
|
||
with torch.no_grad():
|
||
y_pred = self.regenerate(x)
|
||
|
||
y_pred = y_pred.cpu()
|
||
y = batch.cpu() # should already be on CPU but just in case
|
||
rvm = self.rvm(y_pred, y)
|
||
lp.update(**rvm)
|
||
if len(metrics) == 0:
|
||
metrics = rvm
|
||
else:
|
||
for key in rvm.keys():
|
||
metrics[key] = (metrics[key] * n + rvm[key]) / (n + 1)
|
||
metrics = flashy.distrib.average_metrics(metrics)
|
||
return metrics</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Evaluate stage.
|
||
Runs audio reconstruction evaluation.</p></div>
|
||
</dd>
|
||
<dt id="audiocraft.solvers.diffusion.DiffusionSolver.get_condition"><code class="name flex">
|
||
<span>def <span class="ident">get_condition</span></span>(<span>self, wav: torch.Tensor) ‑> torch.Tensor</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">@torch.no_grad()
|
||
def get_condition(self, wav: torch.Tensor) -> torch.Tensor:
|
||
codes, scale = self.codec_model.encode(wav)
|
||
assert scale is None, "Scaled compression models not supported."
|
||
emb = self.codec_model.decode_latent(codes)
|
||
return emb</code></pre>
|
||
</details>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.solvers.diffusion.DiffusionSolver.regenerate"><code class="name flex">
|
||
<span>def <span class="ident">regenerate</span></span>(<span>self, wav: torch.Tensor, step_list: list | None = None)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">@torch.no_grad()
|
||
def regenerate(self, wav: torch.Tensor, step_list: tp.Optional[list] = None):
|
||
"""Regenerate the given waveform."""
|
||
condition = self.get_condition(wav)
|
||
initial = self.schedule.get_initial_noise(self.data_processor.process_data(wav)) # sampling rate changes.
|
||
result = self.schedule.generate_subsampled(self.model, initial=initial, condition=condition,
|
||
step_list=step_list)
|
||
result = self.data_processor.inverse_process(result)
|
||
return result</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Regenerate the given waveform.</p></div>
|
||
</dd>
|
||
</dl>
|
||
<h3>Inherited members</h3>
|
||
<ul class="hlist">
|
||
<li><code><b><a title="audiocraft.solvers.base.StandardSolver" href="base.html#audiocraft.solvers.base.StandardSolver">StandardSolver</a></b></code>:
|
||
<ul class="hlist">
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.autocast" href="base.html#audiocraft.solvers.base.StandardSolver.autocast">autocast</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.best_metric_name" href="base.html#audiocraft.solvers.base.StandardSolver.best_metric_name">best_metric_name</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.commit" href="base.html#audiocraft.solvers.base.StandardSolver.commit">commit</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.common_train_valid" href="base.html#audiocraft.solvers.base.StandardSolver.common_train_valid">common_train_valid</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.generate" href="base.html#audiocraft.solvers.base.StandardSolver.generate">generate</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.get_eval_solver_from_sig" href="base.html#audiocraft.solvers.base.StandardSolver.get_eval_solver_from_sig">get_eval_solver_from_sig</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.initialize_ema" href="base.html#audiocraft.solvers.base.StandardSolver.initialize_ema">initialize_ema</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.load_checkpoints" href="base.html#audiocraft.solvers.base.StandardSolver.load_checkpoints">load_checkpoints</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.log_model_summary" href="base.html#audiocraft.solvers.base.StandardSolver.log_model_summary">log_model_summary</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.register_best_state" href="base.html#audiocraft.solvers.base.StandardSolver.register_best_state">register_best_state</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.register_ema" href="base.html#audiocraft.solvers.base.StandardSolver.register_ema">register_ema</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.restore" href="base.html#audiocraft.solvers.base.StandardSolver.restore">restore</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.run" href="base.html#audiocraft.solvers.base.StandardSolver.run">run</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.run_epoch" href="base.html#audiocraft.solvers.base.StandardSolver.run_epoch">run_epoch</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.run_one_stage" href="base.html#audiocraft.solvers.base.StandardSolver.run_one_stage">run_one_stage</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.run_step" href="base.html#audiocraft.solvers.base.StandardSolver.run_step">run_step</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.save_checkpoints" href="base.html#audiocraft.solvers.base.StandardSolver.save_checkpoints">save_checkpoints</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.should_run_stage" href="base.html#audiocraft.solvers.base.StandardSolver.should_run_stage">should_run_stage</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.should_stop_training" href="base.html#audiocraft.solvers.base.StandardSolver.should_stop_training">should_stop_training</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.show" href="base.html#audiocraft.solvers.base.StandardSolver.show">show</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.train" href="base.html#audiocraft.solvers.base.StandardSolver.train">train</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.update_best_state_from_stage" href="base.html#audiocraft.solvers.base.StandardSolver.update_best_state_from_stage">update_best_state_from_stage</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.base.StandardSolver.valid" href="base.html#audiocraft.solvers.base.StandardSolver.valid">valid</a></code></li>
|
||
</ul>
|
||
</li>
|
||
</ul>
|
||
</dd>
|
||
<dt id="audiocraft.solvers.diffusion.PerStageMetrics"><code class="flex name class">
|
||
<span>class <span class="ident">PerStageMetrics</span></span>
|
||
<span>(</span><span>num_steps: int, num_stages: int = 4)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">class PerStageMetrics:
|
||
"""Handle prompting the metrics per stage.
|
||
It outputs the metrics per range of diffusion states.
|
||
e.g. avg loss when t in [250, 500]
|
||
"""
|
||
def __init__(self, num_steps: int, num_stages: int = 4):
|
||
self.num_steps = num_steps
|
||
self.num_stages = num_stages
|
||
|
||
def __call__(self, losses: dict, step: tp.Union[int, torch.Tensor]):
|
||
if type(step) is int:
|
||
stage = int((step / self.num_steps) * self.num_stages)
|
||
return {f"{name}_{stage}": loss for name, loss in losses.items()}
|
||
elif type(step) is torch.Tensor:
|
||
stage_tensor = ((step / self.num_steps) * self.num_stages).long()
|
||
out: tp.Dict[str, float] = {}
|
||
for stage_idx in range(self.num_stages):
|
||
mask = (stage_tensor == stage_idx)
|
||
N = mask.sum()
|
||
stage_out = {}
|
||
if N > 0: # pass if no elements in the stage
|
||
for name, loss in losses.items():
|
||
stage_loss = (mask * loss).sum() / N
|
||
stage_out[f"{name}_{stage_idx}"] = stage_loss
|
||
out = {**out, **stage_out}
|
||
return out</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Handle prompting the metrics per stage.
|
||
It outputs the metrics per range of diffusion states.
|
||
e.g. avg loss when t in [250, 500]</p></div>
|
||
</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.solvers" href="index.html">audiocraft.solvers</a></code></li>
|
||
</ul>
|
||
</li>
|
||
<li><h3><a href="#header-classes">Classes</a></h3>
|
||
<ul>
|
||
<li>
|
||
<h4><code><a title="audiocraft.solvers.diffusion.DataProcess" href="#audiocraft.solvers.diffusion.DataProcess">DataProcess</a></code></h4>
|
||
<ul class="">
|
||
<li><code><a title="audiocraft.solvers.diffusion.DataProcess.inverse_process" href="#audiocraft.solvers.diffusion.DataProcess.inverse_process">inverse_process</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.diffusion.DataProcess.process_data" href="#audiocraft.solvers.diffusion.DataProcess.process_data">process_data</a></code></li>
|
||
</ul>
|
||
</li>
|
||
<li>
|
||
<h4><code><a title="audiocraft.solvers.diffusion.DiffusionSolver" href="#audiocraft.solvers.diffusion.DiffusionSolver">DiffusionSolver</a></code></h4>
|
||
<ul class="">
|
||
<li><code><a title="audiocraft.solvers.diffusion.DiffusionSolver.build_dataloaders" href="#audiocraft.solvers.diffusion.DiffusionSolver.build_dataloaders">build_dataloaders</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.diffusion.DiffusionSolver.build_model" href="#audiocraft.solvers.diffusion.DiffusionSolver.build_model">build_model</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.diffusion.DiffusionSolver.evaluate" href="#audiocraft.solvers.diffusion.DiffusionSolver.evaluate">evaluate</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.diffusion.DiffusionSolver.get_condition" href="#audiocraft.solvers.diffusion.DiffusionSolver.get_condition">get_condition</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.diffusion.DiffusionSolver.regenerate" href="#audiocraft.solvers.diffusion.DiffusionSolver.regenerate">regenerate</a></code></li>
|
||
</ul>
|
||
</li>
|
||
<li>
|
||
<h4><code><a title="audiocraft.solvers.diffusion.PerStageMetrics" href="#audiocraft.solvers.diffusion.PerStageMetrics">PerStageMetrics</a></code></h4>
|
||
</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>
|