facebookresearch--audiocraft
Alexandre Défossez (autodoc)
691 行
41 KiB
HTML
691 行
41 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.compression 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.compression</code></h1>
|
||
</header>
|
||
<section id="section-intro">
|
||
</section>
|
||
<section>
|
||
</section>
|
||
<section>
|
||
</section>
|
||
<section>
|
||
<h2 class="section-title" id="header-functions">Functions</h2>
|
||
<dl>
|
||
<dt id="audiocraft.solvers.compression.evaluate_audio_reconstruction"><code class="name flex">
|
||
<span>def <span class="ident">evaluate_audio_reconstruction</span></span>(<span>y_pred: torch.Tensor, y: torch.Tensor, cfg: omegaconf.dictconfig.DictConfig) ‑> dict</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def evaluate_audio_reconstruction(y_pred: torch.Tensor, y: torch.Tensor, cfg: omegaconf.DictConfig) -> dict:
|
||
"""Audio reconstruction evaluation method that can be conveniently pickled."""
|
||
metrics = {}
|
||
if cfg.evaluate.metrics.visqol:
|
||
visqol = builders.get_visqol(cfg.metrics.visqol)
|
||
metrics['visqol'] = visqol(y_pred, y, cfg.sample_rate)
|
||
sisnr = builders.get_loss('sisnr', cfg)
|
||
metrics['sisnr'] = sisnr(y_pred, y)
|
||
return metrics</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Audio reconstruction evaluation method that can be conveniently pickled.</p></div>
|
||
</dd>
|
||
</dl>
|
||
</section>
|
||
<section>
|
||
<h2 class="section-title" id="header-classes">Classes</h2>
|
||
<dl>
|
||
<dt id="audiocraft.solvers.compression.CompressionSolver"><code class="flex name class">
|
||
<span>class <span class="ident">CompressionSolver</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 CompressionSolver(base.StandardSolver):
|
||
"""Solver for compression task.
|
||
|
||
The compression task combines a set of perceptual and objective losses
|
||
to train an EncodecModel (composed of an encoder-decoder and a quantizer)
|
||
to perform high fidelity audio reconstruction.
|
||
"""
|
||
def __init__(self, cfg: omegaconf.DictConfig):
|
||
super().__init__(cfg)
|
||
self.rng: torch.Generator # set at each epoch
|
||
self.adv_losses = builders.get_adversarial_losses(self.cfg)
|
||
self.aux_losses = nn.ModuleDict()
|
||
self.info_losses = nn.ModuleDict()
|
||
assert not cfg.fsdp.use, "FSDP not supported by CompressionSolver."
|
||
loss_weights = dict()
|
||
for loss_name, weight in self.cfg.losses.items():
|
||
if loss_name in ['adv', 'feat']:
|
||
for adv_name, _ in self.adv_losses.items():
|
||
loss_weights[f'{loss_name}_{adv_name}'] = weight
|
||
elif weight > 0:
|
||
self.aux_losses[loss_name] = builders.get_loss(loss_name, self.cfg)
|
||
loss_weights[loss_name] = weight
|
||
else:
|
||
self.info_losses[loss_name] = builders.get_loss(loss_name, self.cfg)
|
||
self.balancer = builders.get_balancer(loss_weights, self.cfg.balancer)
|
||
self.register_stateful('adv_losses')
|
||
|
||
@property
|
||
def best_metric_name(self) -> tp.Optional[str]:
|
||
# best model is the last for the compression model
|
||
return None
|
||
|
||
def build_model(self):
|
||
"""Instantiate model and optimizer."""
|
||
# Model and optimizer
|
||
self.model = models.builders.get_compression_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):
|
||
"""Instantiate audio dataloaders for each stage."""
|
||
self.dataloaders = builders.get_audio_datasets(self.cfg)
|
||
|
||
def show(self):
|
||
"""Show the compression model and employed adversarial loss."""
|
||
self.logger.info(f"Compression model with {self.model.quantizer.total_codebooks} codebooks:")
|
||
self.log_model_summary(self.model)
|
||
self.logger.info("Adversarial loss:")
|
||
self.log_model_summary(self.adv_losses)
|
||
self.logger.info("Auxiliary losses:")
|
||
self.logger.info(self.aux_losses)
|
||
self.logger.info("Info losses:")
|
||
self.logger.info(self.info_losses)
|
||
|
||
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)
|
||
y = x.clone()
|
||
|
||
qres = self.model(x)
|
||
assert isinstance(qres, quantization.QuantizedResult)
|
||
y_pred = qres.x
|
||
# Log bandwidth in kb/s
|
||
metrics['bandwidth'] = qres.bandwidth.mean()
|
||
|
||
if self.is_training:
|
||
d_losses: dict = {}
|
||
if len(self.adv_losses) > 0 and torch.rand(1, generator=self.rng).item() <= 1 / self.cfg.adversarial.every:
|
||
for adv_name, adversary in self.adv_losses.items():
|
||
disc_loss = adversary.train_adv(y_pred, y)
|
||
d_losses[f'd_{adv_name}'] = disc_loss
|
||
metrics['d_loss'] = torch.sum(torch.stack(list(d_losses.values())))
|
||
metrics.update(d_losses)
|
||
|
||
balanced_losses: dict = {}
|
||
other_losses: dict = {}
|
||
|
||
# penalty from quantization
|
||
if qres.penalty is not None and qres.penalty.requires_grad:
|
||
other_losses['penalty'] = qres.penalty # penalty term from the quantizer
|
||
|
||
# adversarial losses
|
||
for adv_name, adversary in self.adv_losses.items():
|
||
adv_loss, feat_loss = adversary(y_pred, y)
|
||
balanced_losses[f'adv_{adv_name}'] = adv_loss
|
||
balanced_losses[f'feat_{adv_name}'] = feat_loss
|
||
|
||
# auxiliary losses
|
||
for loss_name, criterion in self.aux_losses.items():
|
||
loss = criterion(y_pred, y)
|
||
balanced_losses[loss_name] = loss
|
||
|
||
# weighted losses
|
||
metrics.update(balanced_losses)
|
||
metrics.update(other_losses)
|
||
metrics.update(qres.metrics)
|
||
|
||
if self.is_training:
|
||
# backprop losses that are not handled by balancer
|
||
other_loss = torch.tensor(0., device=self.device)
|
||
if 'penalty' in other_losses:
|
||
other_loss += other_losses['penalty']
|
||
if other_loss.requires_grad:
|
||
other_loss.backward(retain_graph=True)
|
||
ratio1 = sum(p.grad.data.norm(p=2).pow(2)
|
||
for p in self.model.parameters() if p.grad is not None)
|
||
assert isinstance(ratio1, torch.Tensor)
|
||
metrics['ratio1'] = ratio1.sqrt()
|
||
|
||
# balancer losses backward, returns effective training loss
|
||
# with effective weights at the current batch.
|
||
metrics['g_loss'] = self.balancer.backward(balanced_losses, y_pred)
|
||
# add metrics corresponding to weight ratios
|
||
metrics.update(self.balancer.metrics)
|
||
ratio2 = sum(p.grad.data.norm(p=2).pow(2)
|
||
for p in self.model.parameters() if p.grad is not None)
|
||
assert isinstance(ratio2, torch.Tensor)
|
||
metrics['ratio2'] = ratio2.sqrt()
|
||
|
||
# optim
|
||
flashy.distrib.sync_model(self.model)
|
||
if self.cfg.optim.max_norm:
|
||
torch.nn.utils.clip_grad_norm_(
|
||
self.model.parameters(), self.cfg.optim.max_norm
|
||
)
|
||
self.optimizer.step()
|
||
self.optimizer.zero_grad()
|
||
|
||
# informative losses only
|
||
info_losses: dict = {}
|
||
with torch.no_grad():
|
||
for loss_name, criterion in self.info_losses.items():
|
||
loss = criterion(y_pred, y)
|
||
info_losses[loss_name] = loss
|
||
|
||
metrics.update(info_losses)
|
||
|
||
# aggregated GAN losses: this is useful to report adv and feat across different adversarial loss setups
|
||
adv_losses = [loss for loss_name, loss in metrics.items() if loss_name.startswith('adv')]
|
||
if len(adv_losses) > 0:
|
||
metrics['adv'] = torch.sum(torch.stack(adv_losses))
|
||
feat_losses = [loss for loss_name, loss in metrics.items() if loss_name.startswith('feat')]
|
||
if len(feat_losses) > 0:
|
||
metrics['feat'] = torch.sum(torch.stack(feat_losses))
|
||
|
||
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)
|
||
# run epoch
|
||
super().run_epoch()
|
||
|
||
def evaluate(self):
|
||
"""Evaluate stage. Runs audio reconstruction evaluation."""
|
||
self.model.eval()
|
||
evaluate_stage_name = str(self.current_stage)
|
||
|
||
loader = self.dataloaders['evaluate']
|
||
updates = len(loader)
|
||
lp = self.log_progress(f'{evaluate_stage_name} inference', loader, total=updates, updates=self.log_updates)
|
||
average = flashy.averager()
|
||
|
||
pendings = []
|
||
ctx = multiprocessing.get_context('spawn')
|
||
with get_pool_executor(self.cfg.evaluate.num_workers, mp_context=ctx) as pool:
|
||
for idx, batch in enumerate(lp):
|
||
x = batch.to(self.device)
|
||
with torch.no_grad():
|
||
qres = self.model(x)
|
||
|
||
y_pred = qres.x.cpu()
|
||
y = batch.cpu() # should already be on CPU but just in case
|
||
pendings.append(pool.submit(evaluate_audio_reconstruction, y_pred, y, self.cfg))
|
||
|
||
metrics_lp = self.log_progress(f'{evaluate_stage_name} metrics', pendings, updates=self.log_updates)
|
||
for pending in metrics_lp:
|
||
metrics = pending.result()
|
||
metrics = average(metrics)
|
||
|
||
metrics = flashy.distrib.average_metrics(metrics, len(loader))
|
||
return metrics
|
||
|
||
def generate(self):
|
||
"""Generate stage."""
|
||
self.model.eval()
|
||
sample_manager = SampleManager(self.xp, map_reference_to_sample_id=True)
|
||
generate_stage_name = str(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)
|
||
with torch.no_grad():
|
||
qres = self.model(reference)
|
||
assert isinstance(qres, quantization.QuantizedResult)
|
||
|
||
reference = reference.cpu()
|
||
estimate = qres.x.cpu()
|
||
sample_manager.add_samples(estimate, self.epoch, ground_truth_wavs=reference)
|
||
|
||
flashy.distrib.barrier()
|
||
|
||
def load_from_pretrained(self, name: str) -> dict:
|
||
model = models.CompressionModel.get_pretrained(name)
|
||
if isinstance(model, models.DAC):
|
||
raise RuntimeError("Cannot fine tune a DAC model.")
|
||
elif isinstance(model, models.HFEncodecCompressionModel):
|
||
self.logger.warning('Trying to automatically convert a HuggingFace model '
|
||
'to AudioCraft, this might fail!')
|
||
state = model.model.state_dict()
|
||
new_state = {}
|
||
for k, v in state.items():
|
||
if k.startswith('decoder.layers') and '.conv.' in k and '.block.' not in k:
|
||
# We need to determine if this a convtr or a regular conv.
|
||
layer = int(k.split('.')[2])
|
||
if isinstance(model.model.decoder.layers[layer].conv, torch.nn.ConvTranspose1d):
|
||
|
||
k = k.replace('.conv.', '.convtr.')
|
||
k = k.replace('encoder.layers.', 'encoder.model.')
|
||
k = k.replace('decoder.layers.', 'decoder.model.')
|
||
k = k.replace('conv.', 'conv.conv.')
|
||
k = k.replace('convtr.', 'convtr.convtr.')
|
||
k = k.replace('quantizer.layers.', 'quantizer.vq.layers.')
|
||
k = k.replace('.codebook.', '._codebook.')
|
||
new_state[k] = v
|
||
state = new_state
|
||
elif isinstance(model, models.EncodecModel):
|
||
state = model.state_dict()
|
||
else:
|
||
raise RuntimeError(f"Cannot fine tune model type {type(model)}.")
|
||
return {
|
||
'best_state': {'model': state}
|
||
}
|
||
|
||
@staticmethod
|
||
def model_from_checkpoint(checkpoint_path: tp.Union[Path, str],
|
||
device: tp.Union[torch.device, str] = 'cpu') -> models.CompressionModel:
|
||
"""Instantiate a CompressionModel from a given checkpoint path or dora sig.
|
||
This method is a convenient endpoint to load a CompressionModel to use in other solvers.
|
||
|
||
Args:
|
||
checkpoint_path (Path or str): Path to checkpoint or dora sig from where the checkpoint is resolved.
|
||
This also supports pre-trained models by using a path of the form //pretrained/NAME.
|
||
See `model_from_pretrained` for a list of supported pretrained models.
|
||
use_ema (bool): Use EMA variant of the model instead of the actual model.
|
||
device (torch.device or str): Device on which the model is loaded.
|
||
"""
|
||
checkpoint_path = str(checkpoint_path)
|
||
if checkpoint_path.startswith('//pretrained/'):
|
||
name = checkpoint_path.split('/', 3)[-1]
|
||
return models.CompressionModel.get_pretrained(name, device)
|
||
logger = logging.getLogger(__name__)
|
||
logger.info(f"Loading compression model from checkpoint: {checkpoint_path}")
|
||
_checkpoint_path = checkpoint.resolve_checkpoint_path(checkpoint_path, use_fsdp=False)
|
||
assert _checkpoint_path is not None, f"Could not resolve compression model checkpoint path: {checkpoint_path}"
|
||
state = checkpoint.load_checkpoint(_checkpoint_path)
|
||
assert state is not None and 'xp.cfg' in state, f"Could not load compression model from ckpt: {checkpoint_path}"
|
||
cfg = state['xp.cfg']
|
||
cfg.device = device
|
||
compression_model = models.builders.get_compression_model(cfg).to(device)
|
||
assert compression_model.sample_rate == cfg.sample_rate, "Compression model sample rate should match"
|
||
|
||
assert 'best_state' in state and state['best_state'] != {}
|
||
assert 'exported' not in state, "When loading an exported checkpoint, use the //pretrained/ prefix."
|
||
compression_model.load_state_dict(state['best_state']['model'])
|
||
compression_model.eval()
|
||
logger.info("Compression model loaded!")
|
||
return compression_model
|
||
|
||
@staticmethod
|
||
def wrapped_model_from_checkpoint(cfg: omegaconf.DictConfig,
|
||
checkpoint_path: tp.Union[Path, str],
|
||
device: tp.Union[torch.device, str] = 'cpu') -> models.CompressionModel:
|
||
"""Instantiate a wrapped CompressionModel from a given checkpoint path or dora sig.
|
||
|
||
Args:
|
||
cfg (omegaconf.DictConfig): Configuration to read from for wrapped mode.
|
||
checkpoint_path (Path or str): Path to checkpoint or dora sig from where the checkpoint is resolved.
|
||
use_ema (bool): Use EMA variant of the model instead of the actual model.
|
||
device (torch.device or str): Device on which the model is loaded.
|
||
"""
|
||
compression_model = CompressionSolver.model_from_checkpoint(checkpoint_path, device)
|
||
compression_model = models.builders.get_wrapped_compression_model(compression_model, cfg)
|
||
return compression_model</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Solver for compression task.</p>
|
||
<p>The compression task combines a set of perceptual and objective losses
|
||
to train an EncodecModel (composed of an encoder-decoder and a quantizer)
|
||
to perform high fidelity audio reconstruction.</p></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>Static methods</h3>
|
||
<dl>
|
||
<dt id="audiocraft.solvers.compression.CompressionSolver.model_from_checkpoint"><code class="name flex">
|
||
<span>def <span class="ident">model_from_checkpoint</span></span>(<span>checkpoint_path: str | pathlib.Path, device: torch.device | str = 'cpu') ‑> <a title="audiocraft.models.encodec.CompressionModel" href="../models/encodec.html#audiocraft.models.encodec.CompressionModel">CompressionModel</a></span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">@staticmethod
|
||
def model_from_checkpoint(checkpoint_path: tp.Union[Path, str],
|
||
device: tp.Union[torch.device, str] = 'cpu') -> models.CompressionModel:
|
||
"""Instantiate a CompressionModel from a given checkpoint path or dora sig.
|
||
This method is a convenient endpoint to load a CompressionModel to use in other solvers.
|
||
|
||
Args:
|
||
checkpoint_path (Path or str): Path to checkpoint or dora sig from where the checkpoint is resolved.
|
||
This also supports pre-trained models by using a path of the form //pretrained/NAME.
|
||
See `model_from_pretrained` for a list of supported pretrained models.
|
||
use_ema (bool): Use EMA variant of the model instead of the actual model.
|
||
device (torch.device or str): Device on which the model is loaded.
|
||
"""
|
||
checkpoint_path = str(checkpoint_path)
|
||
if checkpoint_path.startswith('//pretrained/'):
|
||
name = checkpoint_path.split('/', 3)[-1]
|
||
return models.CompressionModel.get_pretrained(name, device)
|
||
logger = logging.getLogger(__name__)
|
||
logger.info(f"Loading compression model from checkpoint: {checkpoint_path}")
|
||
_checkpoint_path = checkpoint.resolve_checkpoint_path(checkpoint_path, use_fsdp=False)
|
||
assert _checkpoint_path is not None, f"Could not resolve compression model checkpoint path: {checkpoint_path}"
|
||
state = checkpoint.load_checkpoint(_checkpoint_path)
|
||
assert state is not None and 'xp.cfg' in state, f"Could not load compression model from ckpt: {checkpoint_path}"
|
||
cfg = state['xp.cfg']
|
||
cfg.device = device
|
||
compression_model = models.builders.get_compression_model(cfg).to(device)
|
||
assert compression_model.sample_rate == cfg.sample_rate, "Compression model sample rate should match"
|
||
|
||
assert 'best_state' in state and state['best_state'] != {}
|
||
assert 'exported' not in state, "When loading an exported checkpoint, use the //pretrained/ prefix."
|
||
compression_model.load_state_dict(state['best_state']['model'])
|
||
compression_model.eval()
|
||
logger.info("Compression model loaded!")
|
||
return compression_model</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Instantiate a CompressionModel from a given checkpoint path or dora sig.
|
||
This method is a convenient endpoint to load a CompressionModel to use in other solvers.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>checkpoint_path</code></strong> : <code>Path</code> or <code>str</code></dt>
|
||
<dd>Path to checkpoint or dora sig from where the checkpoint is resolved.
|
||
This also supports pre-trained models by using a path of the form //pretrained/NAME.
|
||
See <code>model_from_pretrained</code> for a list of supported pretrained models.</dd>
|
||
<dt><strong><code>use_ema</code></strong> : <code>bool</code></dt>
|
||
<dd>Use EMA variant of the model instead of the actual model.</dd>
|
||
<dt><strong><code>device</code></strong> : <code>torch.device</code> or <code>str</code></dt>
|
||
<dd>Device on which the model is loaded.</dd>
|
||
</dl></div>
|
||
</dd>
|
||
<dt id="audiocraft.solvers.compression.CompressionSolver.wrapped_model_from_checkpoint"><code class="name flex">
|
||
<span>def <span class="ident">wrapped_model_from_checkpoint</span></span>(<span>cfg: omegaconf.dictconfig.DictConfig,<br>checkpoint_path: str | pathlib.Path,<br>device: torch.device | str = 'cpu') ‑> <a title="audiocraft.models.encodec.CompressionModel" href="../models/encodec.html#audiocraft.models.encodec.CompressionModel">CompressionModel</a></span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">@staticmethod
|
||
def wrapped_model_from_checkpoint(cfg: omegaconf.DictConfig,
|
||
checkpoint_path: tp.Union[Path, str],
|
||
device: tp.Union[torch.device, str] = 'cpu') -> models.CompressionModel:
|
||
"""Instantiate a wrapped CompressionModel from a given checkpoint path or dora sig.
|
||
|
||
Args:
|
||
cfg (omegaconf.DictConfig): Configuration to read from for wrapped mode.
|
||
checkpoint_path (Path or str): Path to checkpoint or dora sig from where the checkpoint is resolved.
|
||
use_ema (bool): Use EMA variant of the model instead of the actual model.
|
||
device (torch.device or str): Device on which the model is loaded.
|
||
"""
|
||
compression_model = CompressionSolver.model_from_checkpoint(checkpoint_path, device)
|
||
compression_model = models.builders.get_wrapped_compression_model(compression_model, cfg)
|
||
return compression_model</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Instantiate a wrapped CompressionModel from a given checkpoint path or dora sig.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>cfg</code></strong> : <code>omegaconf.DictConfig</code></dt>
|
||
<dd>Configuration to read from for wrapped mode.</dd>
|
||
<dt><strong><code>checkpoint_path</code></strong> : <code>Path</code> or <code>str</code></dt>
|
||
<dd>Path to checkpoint or dora sig from where the checkpoint is resolved.</dd>
|
||
<dt><strong><code>use_ema</code></strong> : <code>bool</code></dt>
|
||
<dd>Use EMA variant of the model instead of the actual model.</dd>
|
||
<dt><strong><code>device</code></strong> : <code>torch.device</code> or <code>str</code></dt>
|
||
<dd>Device on which the model is loaded.</dd>
|
||
</dl></div>
|
||
</dd>
|
||
</dl>
|
||
<h3>Methods</h3>
|
||
<dl>
|
||
<dt id="audiocraft.solvers.compression.CompressionSolver.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):
|
||
"""Instantiate audio dataloaders for each stage."""
|
||
self.dataloaders = builders.get_audio_datasets(self.cfg)</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Instantiate audio dataloaders for each stage.</p></div>
|
||
</dd>
|
||
<dt id="audiocraft.solvers.compression.CompressionSolver.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):
|
||
"""Instantiate model and optimizer."""
|
||
# Model and optimizer
|
||
self.model = models.builders.get_compression_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>Instantiate model and optimizer.</p></div>
|
||
</dd>
|
||
<dt id="audiocraft.solvers.compression.CompressionSolver.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 = str(self.current_stage)
|
||
|
||
loader = self.dataloaders['evaluate']
|
||
updates = len(loader)
|
||
lp = self.log_progress(f'{evaluate_stage_name} inference', loader, total=updates, updates=self.log_updates)
|
||
average = flashy.averager()
|
||
|
||
pendings = []
|
||
ctx = multiprocessing.get_context('spawn')
|
||
with get_pool_executor(self.cfg.evaluate.num_workers, mp_context=ctx) as pool:
|
||
for idx, batch in enumerate(lp):
|
||
x = batch.to(self.device)
|
||
with torch.no_grad():
|
||
qres = self.model(x)
|
||
|
||
y_pred = qres.x.cpu()
|
||
y = batch.cpu() # should already be on CPU but just in case
|
||
pendings.append(pool.submit(evaluate_audio_reconstruction, y_pred, y, self.cfg))
|
||
|
||
metrics_lp = self.log_progress(f'{evaluate_stage_name} metrics', pendings, updates=self.log_updates)
|
||
for pending in metrics_lp:
|
||
metrics = pending.result()
|
||
metrics = average(metrics)
|
||
|
||
metrics = flashy.distrib.average_metrics(metrics, len(loader))
|
||
return metrics</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Evaluate stage. Runs audio reconstruction evaluation.</p></div>
|
||
</dd>
|
||
<dt id="audiocraft.solvers.compression.CompressionSolver.load_from_pretrained"><code class="name flex">
|
||
<span>def <span class="ident">load_from_pretrained</span></span>(<span>self, name: str) ‑> dict</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def load_from_pretrained(self, name: str) -> dict:
|
||
model = models.CompressionModel.get_pretrained(name)
|
||
if isinstance(model, models.DAC):
|
||
raise RuntimeError("Cannot fine tune a DAC model.")
|
||
elif isinstance(model, models.HFEncodecCompressionModel):
|
||
self.logger.warning('Trying to automatically convert a HuggingFace model '
|
||
'to AudioCraft, this might fail!')
|
||
state = model.model.state_dict()
|
||
new_state = {}
|
||
for k, v in state.items():
|
||
if k.startswith('decoder.layers') and '.conv.' in k and '.block.' not in k:
|
||
# We need to determine if this a convtr or a regular conv.
|
||
layer = int(k.split('.')[2])
|
||
if isinstance(model.model.decoder.layers[layer].conv, torch.nn.ConvTranspose1d):
|
||
|
||
k = k.replace('.conv.', '.convtr.')
|
||
k = k.replace('encoder.layers.', 'encoder.model.')
|
||
k = k.replace('decoder.layers.', 'decoder.model.')
|
||
k = k.replace('conv.', 'conv.conv.')
|
||
k = k.replace('convtr.', 'convtr.convtr.')
|
||
k = k.replace('quantizer.layers.', 'quantizer.vq.layers.')
|
||
k = k.replace('.codebook.', '._codebook.')
|
||
new_state[k] = v
|
||
state = new_state
|
||
elif isinstance(model, models.EncodecModel):
|
||
state = model.state_dict()
|
||
else:
|
||
raise RuntimeError(f"Cannot fine tune model type {type(model)}.")
|
||
return {
|
||
'best_state': {'model': state}
|
||
}</code></pre>
|
||
</details>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.solvers.compression.CompressionSolver.show"><code class="name flex">
|
||
<span>def <span class="ident">show</span></span>(<span>self)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def show(self):
|
||
"""Show the compression model and employed adversarial loss."""
|
||
self.logger.info(f"Compression model with {self.model.quantizer.total_codebooks} codebooks:")
|
||
self.log_model_summary(self.model)
|
||
self.logger.info("Adversarial loss:")
|
||
self.log_model_summary(self.adv_losses)
|
||
self.logger.info("Auxiliary losses:")
|
||
self.logger.info(self.aux_losses)
|
||
self.logger.info("Info losses:")
|
||
self.logger.info(self.info_losses)</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Show the compression model and employed adversarial loss.</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.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>
|
||
</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-functions">Functions</a></h3>
|
||
<ul class="">
|
||
<li><code><a title="audiocraft.solvers.compression.evaluate_audio_reconstruction" href="#audiocraft.solvers.compression.evaluate_audio_reconstruction">evaluate_audio_reconstruction</a></code></li>
|
||
</ul>
|
||
</li>
|
||
<li><h3><a href="#header-classes">Classes</a></h3>
|
||
<ul>
|
||
<li>
|
||
<h4><code><a title="audiocraft.solvers.compression.CompressionSolver" href="#audiocraft.solvers.compression.CompressionSolver">CompressionSolver</a></code></h4>
|
||
<ul class="">
|
||
<li><code><a title="audiocraft.solvers.compression.CompressionSolver.build_dataloaders" href="#audiocraft.solvers.compression.CompressionSolver.build_dataloaders">build_dataloaders</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.compression.CompressionSolver.build_model" href="#audiocraft.solvers.compression.CompressionSolver.build_model">build_model</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.compression.CompressionSolver.evaluate" href="#audiocraft.solvers.compression.CompressionSolver.evaluate">evaluate</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.compression.CompressionSolver.load_from_pretrained" href="#audiocraft.solvers.compression.CompressionSolver.load_from_pretrained">load_from_pretrained</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.compression.CompressionSolver.model_from_checkpoint" href="#audiocraft.solvers.compression.CompressionSolver.model_from_checkpoint">model_from_checkpoint</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.compression.CompressionSolver.show" href="#audiocraft.solvers.compression.CompressionSolver.show">show</a></code></li>
|
||
<li><code><a title="audiocraft.solvers.compression.CompressionSolver.wrapped_model_from_checkpoint" href="#audiocraft.solvers.compression.CompressionSolver.wrapped_model_from_checkpoint">wrapped_model_from_checkpoint</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>
|