facebookresearch--audiocraft
Alexandre Défossez (autodoc)
733 行
35 KiB
HTML
733 行
35 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.utils.utils 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.utils.utils</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.utils.utils.collate"><code class="name flex">
|
||
<span>def <span class="ident">collate</span></span>(<span>tensors: List[torch.Tensor], dim: int = 0) ‑> Tuple[torch.Tensor, torch.Tensor]</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def collate(tensors: tp.List[torch.Tensor], dim: int = 0) -> tp.Tuple[torch.Tensor, torch.Tensor]:
|
||
"""Get a list of tensors and collate them to a single tensor. according to the following logic:
|
||
- `dim` specifies the time dimension which will be stacked and padded.
|
||
- The output will contain 1 new dimension (dimension index 0) which will be the size of
|
||
of the original list.
|
||
|
||
Args:
|
||
tensors (tp.List[torch.Tensor]): List of tensors to collate.
|
||
dim (int): Dimension which will be stacked and padded.
|
||
Returns:
|
||
tp.Tuple[torch.Tensor, torch.Tensor]:
|
||
torch.Tensor: Stacked and padded tensor. The output will contain 1 new dimension
|
||
(dimension index 0) which will be the size of the original list.
|
||
torch.Tensor: Tensor containing length of original tensor sizes (without padding).
|
||
"""
|
||
tensors = [x.transpose(0, dim) for x in tensors]
|
||
lens = torch.LongTensor([len(x) for x in tensors])
|
||
padded_tensors = pad_sequence(tensors)
|
||
padded_tensors = padded_tensors.transpose(0, 1)
|
||
padded_tensors = padded_tensors.transpose(1, dim + 1)
|
||
return padded_tensors, lens</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Get a list of tensors and collate them to a single tensor. according to the following logic:
|
||
- <code>dim</code> specifies the time dimension which will be stacked and padded.
|
||
- The output will contain 1 new dimension (dimension index 0) which will be the size of
|
||
of the original list.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>tensors</code></strong> : <code>tp.List[torch.Tensor]</code></dt>
|
||
<dd>List of tensors to collate.</dd>
|
||
<dt><strong><code>dim</code></strong> : <code>int</code></dt>
|
||
<dd>Dimension which will be stacked and padded.</dd>
|
||
</dl>
|
||
<h2 id="returns">Returns</h2>
|
||
<dl>
|
||
<dt>tp.Tuple[torch.Tensor, torch.Tensor]:</dt>
|
||
<dt><code>
|
||
torch.Tensor</code></dt>
|
||
<dd>Stacked and padded tensor. The output will contain 1 new dimension
|
||
(dimension index 0) which will be the size of the original list.
|
||
torch.Tensor: Tensor containing length of original tensor sizes (without padding).</dd>
|
||
</dl></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.construct_frame_chords"><code class="name flex">
|
||
<span>def <span class="ident">construct_frame_chords</span></span>(<span>min_timestamp: int,<br>chord_changes: List[Tuple[float, str]],<br>mapping_dict: Dict,<br>prev_chord: str,<br>frame_rate: float,<br>segment_duration: float) ‑> List[str]</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def construct_frame_chords(
|
||
min_timestamp: int,
|
||
chord_changes: tp.List[tp.Tuple[float, str]],
|
||
mapping_dict: tp.Dict,
|
||
prev_chord: str,
|
||
frame_rate: float,
|
||
segment_duration: float,
|
||
) -> tp.List[str]:
|
||
""" Translate symbolic chords [(start_time, tuples),...] into a frame-level int sequence"""
|
||
|
||
frames = [
|
||
frame / frame_rate
|
||
for frame in range(
|
||
min_timestamp, int(min_timestamp + segment_duration * frame_rate)
|
||
)
|
||
]
|
||
|
||
frame_chords = []
|
||
current_chord = prev_chord
|
||
|
||
for frame in frames:
|
||
while chord_changes and frame >= chord_changes[0][0]:
|
||
current_chord = chord_changes.pop(0)[1]
|
||
current_chord = 'N' if current_chord in {None, ''} else current_chord
|
||
frame_chords.append(mapping_dict[current_chord])
|
||
|
||
return frame_chords</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Translate symbolic chords [(start_time, tuples),…] into a frame-level int sequence</p></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.copy_state"><code class="name flex">
|
||
<span>def <span class="ident">copy_state</span></span>(<span>state: Any,<br>device: torch.device | str = 'cpu',<br>dtype: torch.dtype | None = None) ‑> Any</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def copy_state(state: tp.Any, device: tp.Union[torch.device, str] = 'cpu',
|
||
dtype: tp.Optional[torch.dtype] = None) -> tp.Any:
|
||
if isinstance(state, torch.Tensor):
|
||
if dtype is None or not state.is_floating_point():
|
||
dtype = state.dtype
|
||
return state.detach().to(device=device, dtype=dtype, copy=True)
|
||
elif isinstance(state, dict):
|
||
return {k: copy_state(v, device, dtype) for k, v in state.items()}
|
||
elif isinstance(state, list):
|
||
return [copy_state(v, device, dtype) for v in state]</code></pre>
|
||
</details>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.dict_from_config"><code class="name flex">
|
||
<span>def <span class="ident">dict_from_config</span></span>(<span>cfg: omegaconf.dictconfig.DictConfig) ‑> dict</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def dict_from_config(cfg: omegaconf.DictConfig) -> dict:
|
||
"""Convenience function to map an omegaconf configuration to a dictionary.
|
||
|
||
Args:
|
||
cfg (omegaconf.DictConfig): Original configuration to map to dict.
|
||
Returns:
|
||
dict: Config as dictionary object.
|
||
"""
|
||
dct = omegaconf.OmegaConf.to_container(cfg, resolve=True)
|
||
assert isinstance(dct, dict)
|
||
return dct</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Convenience function to map an omegaconf configuration to a dictionary.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>cfg</code></strong> : <code>omegaconf.DictConfig</code></dt>
|
||
<dd>Original configuration to map to dict.</dd>
|
||
</dl>
|
||
<h2 id="returns">Returns</h2>
|
||
<dl>
|
||
<dt><code>dict</code></dt>
|
||
<dd>Config as dictionary object.</dd>
|
||
</dl></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.get_dataset_from_loader"><code class="name flex">
|
||
<span>def <span class="ident">get_dataset_from_loader</span></span>(<span>dataloader)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def get_dataset_from_loader(dataloader):
|
||
dataset = dataloader.dataset
|
||
if isinstance(dataset, torch.utils.data.Subset):
|
||
return dataset.dataset
|
||
else:
|
||
return dataset</code></pre>
|
||
</details>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.get_loader"><code class="name flex">
|
||
<span>def <span class="ident">get_loader</span></span>(<span>dataset,<br>num_samples: int | None,<br>batch_size: int,<br>num_workers: int,<br>seed: int,<br>**kwargs) ‑> torch.utils.data.dataloader.DataLoader</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def get_loader(dataset, num_samples: tp.Optional[int], batch_size: int,
|
||
num_workers: int, seed: int, **kwargs) -> torch.utils.data.DataLoader:
|
||
"""Convenience function to load dataset into a dataloader with optional subset sampling.
|
||
|
||
Args:
|
||
dataset: Dataset to load.
|
||
num_samples (Optional[int]): Number of samples to limit subset size.
|
||
batch_size (int): Batch size.
|
||
num_workers (int): Number of workers for data loading.
|
||
seed (int): Random seed.
|
||
"""
|
||
if num_samples is not None:
|
||
dataset = random_subset(dataset, num_samples, seed)
|
||
|
||
dataloader = flashy.distrib.loader(
|
||
dataset,
|
||
batch_size=batch_size,
|
||
num_workers=num_workers,
|
||
**kwargs
|
||
)
|
||
return dataloader</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Convenience function to load dataset into a dataloader with optional subset sampling.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>dataset</code></strong></dt>
|
||
<dd>Dataset to load.</dd>
|
||
<dt><strong><code>num_samples</code></strong> : <code>Optional[int]</code></dt>
|
||
<dd>Number of samples to limit subset size.</dd>
|
||
<dt><strong><code>batch_size</code></strong> : <code>int</code></dt>
|
||
<dd>Batch size.</dd>
|
||
<dt><strong><code>num_workers</code></strong> : <code>int</code></dt>
|
||
<dd>Number of workers for data loading.</dd>
|
||
<dt><strong><code>seed</code></strong> : <code>int</code></dt>
|
||
<dd>Random seed.</dd>
|
||
</dl></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.get_pool_executor"><code class="name flex">
|
||
<span>def <span class="ident">get_pool_executor</span></span>(<span>num_workers: int, mp_context=None)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def get_pool_executor(num_workers: int, mp_context=None):
|
||
return ProcessPoolExecutor(num_workers, mp_context) if num_workers > 1 else DummyPoolExecutor(1)</code></pre>
|
||
</details>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.hash_trick"><code class="name flex">
|
||
<span>def <span class="ident">hash_trick</span></span>(<span>word: str, vocab_size: int) ‑> int</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def hash_trick(word: str, vocab_size: int) -> int:
|
||
"""Hash trick to pair each word with an index
|
||
|
||
Args:
|
||
word (str): word we wish to convert to an index
|
||
vocab_size (int): size of the vocabulary
|
||
Returns:
|
||
int: index of the word in the embedding LUT
|
||
"""
|
||
hash = int(hashlib.sha256(word.encode("utf-8")).hexdigest(), 16)
|
||
return hash % vocab_size</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Hash trick to pair each word with an index</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>word</code></strong> : <code>str</code></dt>
|
||
<dd>word we wish to convert to an index</dd>
|
||
<dt><strong><code>vocab_size</code></strong> : <code>int</code></dt>
|
||
<dd>size of the vocabulary</dd>
|
||
</dl>
|
||
<h2 id="returns">Returns</h2>
|
||
<dl>
|
||
<dt><code>int</code></dt>
|
||
<dd>index of the word in the embedding LUT</dd>
|
||
</dl></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.is_jsonable"><code class="name flex">
|
||
<span>def <span class="ident">is_jsonable</span></span>(<span>x: Any)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def is_jsonable(x: tp.Any):
|
||
"""Check if an object can be serialized into a json:"""
|
||
try:
|
||
json.dumps(x)
|
||
return True
|
||
except (TypeError, OverflowError):
|
||
return False</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Check if an object can be serialized into a json:</p></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.length_to_mask"><code class="name flex">
|
||
<span>def <span class="ident">length_to_mask</span></span>(<span>lengths: torch.Tensor, max_len: int | None = None) ‑> torch.Tensor</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def length_to_mask(lengths: torch.Tensor, max_len: tp.Optional[int] = None) -> torch.Tensor:
|
||
"""Utility function to convert a tensor of sequence lengths to a mask (useful when working on padded sequences).
|
||
For example: [3, 5] => [[1, 1, 1, 0, 0], [1, 1, 1, 1, 1]]
|
||
|
||
Args:
|
||
lengths (torch.Tensor): tensor with lengths
|
||
max_len (int): can set the max length manually. Defaults to None.
|
||
Returns:
|
||
torch.Tensor: mask with 0s where there is pad tokens else 1s
|
||
"""
|
||
assert len(lengths.shape) == 1, "Length shape should be 1 dimensional."
|
||
final_length = lengths.max().item() if not max_len else max_len
|
||
final_length = max(final_length, 1) # if all seqs are of len zero we don't want a zero-size tensor
|
||
return torch.arange(final_length, device=lengths.device)[None, :] < lengths[:, None]</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Utility function to convert a tensor of sequence lengths to a mask (useful when working on padded sequences).
|
||
For example: [3, 5] => [[1, 1, 1, 0, 0], [1, 1, 1, 1, 1]]</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>lengths</code></strong> : <code>torch.Tensor</code></dt>
|
||
<dd>tensor with lengths</dd>
|
||
<dt><strong><code>max_len</code></strong> : <code>int</code></dt>
|
||
<dd>can set the max length manually. Defaults to None.</dd>
|
||
</dl>
|
||
<h2 id="returns">Returns</h2>
|
||
<dl>
|
||
<dt><code>torch.Tensor</code></dt>
|
||
<dd>mask with 0s where there is pad tokens else 1s</dd>
|
||
</dl></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.load_clap_state_dict"><code class="name flex">
|
||
<span>def <span class="ident">load_clap_state_dict</span></span>(<span>clap_model, path: str | pathlib.Path)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def load_clap_state_dict(clap_model, path: tp.Union[str, Path]):
|
||
"""Wrapper around state dict loading of CLAP model
|
||
addressing compatibility issues between CLAP and AudioCraft
|
||
HuggingFace transformer version.
|
||
See: https://github.com/LAION-AI/CLAP/issues/118
|
||
"""
|
||
from clap_module.factory import load_state_dict # type: ignore
|
||
pkg = load_state_dict(path)
|
||
pkg.pop('text_branch.embeddings.position_ids', None)
|
||
clap_model.model.load_state_dict(pkg)</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Wrapper around state dict loading of CLAP model
|
||
addressing compatibility issues between CLAP and AudioCraft
|
||
HuggingFace transformer version.
|
||
See: <a href="https://github.com/LAION-AI/CLAP/issues/118">https://github.com/LAION-AI/CLAP/issues/118</a></p></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.model_hash"><code class="name flex">
|
||
<span>def <span class="ident">model_hash</span></span>(<span>model: torch.nn.modules.module.Module) ‑> str</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def model_hash(model: torch.nn.Module) -> str:
|
||
"""Return a model hash. This should allow us to track regressions in model init
|
||
from the logs of past experiments.
|
||
"""
|
||
hasher = hashlib.sha1()
|
||
for p in model.parameters():
|
||
hasher.update(p.data.cpu().numpy().tobytes())
|
||
return hasher.hexdigest()</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Return a model hash. This should allow us to track regressions in model init
|
||
from the logs of past experiments.</p></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.multinomial"><code class="name flex">
|
||
<span>def <span class="ident">multinomial</span></span>(<span>input: torch.Tensor, num_samples: int, replacement=False, *, generator=None)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def multinomial(input: torch.Tensor, num_samples: int, replacement=False, *, generator=None):
|
||
"""torch.multinomial with arbitrary number of dimensions, and number of candidates on the last dimension.
|
||
|
||
Args:
|
||
input (torch.Tensor): The input tensor containing probabilities.
|
||
num_samples (int): Number of samples to draw.
|
||
replacement (bool): Whether to draw with replacement or not.
|
||
Keywords args:
|
||
generator (torch.Generator): A pseudorandom number generator for sampling.
|
||
Returns:
|
||
torch.Tensor: Last dimension contains num_samples indices
|
||
sampled from the multinomial probability distribution
|
||
located in the last dimension of tensor input.
|
||
"""
|
||
input_ = input.reshape(-1, input.shape[-1])
|
||
output_ = torch.multinomial(input_, num_samples=num_samples, replacement=replacement, generator=generator)
|
||
output = output_.reshape(*list(input.shape[:-1]), -1)
|
||
return output</code></pre>
|
||
</details>
|
||
<div class="desc"><p>torch.multinomial with arbitrary number of dimensions, and number of candidates on the last dimension.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>input</code></strong> : <code>torch.Tensor</code></dt>
|
||
<dd>The input tensor containing probabilities.</dd>
|
||
<dt><strong><code>num_samples</code></strong> : <code>int</code></dt>
|
||
<dd>Number of samples to draw.</dd>
|
||
<dt><strong><code>replacement</code></strong> : <code>bool</code></dt>
|
||
<dd>Whether to draw with replacement or not.</dd>
|
||
</dl>
|
||
<p>Keywords args:
|
||
generator (torch.Generator): A pseudorandom number generator for sampling.</p>
|
||
<h2 id="returns">Returns</h2>
|
||
<dl>
|
||
<dt><code>torch.Tensor</code></dt>
|
||
<dd>Last dimension contains num_samples indices
|
||
sampled from the multinomial probability distribution
|
||
located in the last dimension of tensor input.</dd>
|
||
</dl></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.random_subset"><code class="name flex">
|
||
<span>def <span class="ident">random_subset</span></span>(<span>dataset, max_samples: int, seed: int = 42) ‑> torch.utils.data.dataset.Subset</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def random_subset(dataset, max_samples: int, seed: int = 42) -> torch.utils.data.Subset:
|
||
if max_samples >= len(dataset):
|
||
return dataset
|
||
|
||
generator = torch.Generator().manual_seed(seed)
|
||
perm = torch.randperm(len(dataset), generator=generator)
|
||
return torch.utils.data.Subset(dataset, perm[:max_samples].tolist())</code></pre>
|
||
</details>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.sample_top_k"><code class="name flex">
|
||
<span>def <span class="ident">sample_top_k</span></span>(<span>probs: torch.Tensor, k: int) ‑> torch.Tensor</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def sample_top_k(probs: torch.Tensor, k: int) -> torch.Tensor:
|
||
"""Sample next token from top K values along the last dimension of the input probs tensor.
|
||
|
||
Args:
|
||
probs (torch.Tensor): Input probabilities with token candidates on the last dimension.
|
||
k (int): The k in “top-k”.
|
||
Returns:
|
||
torch.Tensor: Sampled tokens.
|
||
"""
|
||
top_k_value, _ = torch.topk(probs, k, dim=-1)
|
||
min_value_top_k = top_k_value[..., [-1]]
|
||
probs *= (probs >= min_value_top_k).float()
|
||
probs.div_(probs.sum(dim=-1, keepdim=True))
|
||
next_token = multinomial(probs, num_samples=1)
|
||
return next_token</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Sample next token from top K values along the last dimension of the input probs tensor.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>probs</code></strong> : <code>torch.Tensor</code></dt>
|
||
<dd>Input probabilities with token candidates on the last dimension.</dd>
|
||
<dt><strong><code>k</code></strong> : <code>int</code></dt>
|
||
<dd>The k in “top-k”.</dd>
|
||
</dl>
|
||
<h2 id="returns">Returns</h2>
|
||
<dl>
|
||
<dt><code>torch.Tensor</code></dt>
|
||
<dd>Sampled tokens.</dd>
|
||
</dl></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.sample_top_p"><code class="name flex">
|
||
<span>def <span class="ident">sample_top_p</span></span>(<span>probs: torch.Tensor, p: float) ‑> torch.Tensor</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def sample_top_p(probs: torch.Tensor, p: float) -> torch.Tensor:
|
||
"""Sample next token from top P probabilities along the last dimension of the input probs tensor.
|
||
|
||
Args:
|
||
probs (torch.Tensor): Input probabilities with token candidates on the last dimension.
|
||
p (int): The p in “top-p”.
|
||
Returns:
|
||
torch.Tensor: Sampled tokens.
|
||
"""
|
||
probs_sort, probs_idx = torch.sort(probs, dim=-1, descending=True)
|
||
probs_sum = torch.cumsum(probs_sort, dim=-1)
|
||
mask = probs_sum - probs_sort > p
|
||
probs_sort *= (~mask).float()
|
||
probs_sort.div_(probs_sort.sum(dim=-1, keepdim=True))
|
||
next_token = multinomial(probs_sort, num_samples=1)
|
||
next_token = torch.gather(probs_idx, -1, next_token)
|
||
return next_token</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Sample next token from top P probabilities along the last dimension of the input probs tensor.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>probs</code></strong> : <code>torch.Tensor</code></dt>
|
||
<dd>Input probabilities with token candidates on the last dimension.</dd>
|
||
<dt><strong><code>p</code></strong> : <code>int</code></dt>
|
||
<dd>The p in “top-p”.</dd>
|
||
</dl>
|
||
<h2 id="returns">Returns</h2>
|
||
<dl>
|
||
<dt><code>torch.Tensor</code></dt>
|
||
<dd>Sampled tokens.</dd>
|
||
</dl></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.swap_state"><code class="name flex">
|
||
<span>def <span class="ident">swap_state</span></span>(<span>model, state, **kwargs)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">@contextmanager
|
||
def swap_state(model, state, **kwargs):
|
||
old_state = copy_state(model.state_dict())
|
||
model.load_state_dict(state, **kwargs)
|
||
try:
|
||
yield
|
||
finally:
|
||
model.load_state_dict(old_state)</code></pre>
|
||
</details>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.warn_once"><code class="name flex">
|
||
<span>def <span class="ident">warn_once</span></span>(<span>logger, msg)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">@lru_cache(None)
|
||
def warn_once(logger, msg):
|
||
"""Warn about a given message only once."""
|
||
logger.warning(msg)</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Warn about a given message only once.</p></div>
|
||
</dd>
|
||
<dt id="audiocraft.utils.utils.with_rank_rng"><code class="name flex">
|
||
<span>def <span class="ident">with_rank_rng</span></span>(<span>base_seed: int = 1234)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def with_rank_rng(base_seed: int = 1234):
|
||
"""Decorator for a function so that the function will use a Random Number Generator
|
||
whose state depend on the GPU rank. The original RNG state is restored upon returning.
|
||
|
||
Args:
|
||
base_seed (int): Random seed.
|
||
"""
|
||
def _decorator(fun: tp.Callable):
|
||
@wraps(fun)
|
||
def _decorated(*args, **kwargs):
|
||
state = torch.get_rng_state()
|
||
seed = base_seed ^ flashy.distrib.rank()
|
||
torch.manual_seed(seed)
|
||
logger.debug('Rank dependent seed set to %d', seed)
|
||
try:
|
||
return fun(*args, **kwargs)
|
||
finally:
|
||
torch.set_rng_state(state)
|
||
logger.debug('RNG state restored.')
|
||
return _decorated
|
||
return _decorator</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Decorator for a function so that the function will use a Random Number Generator
|
||
whose state depend on the GPU rank. The original RNG state is restored upon returning.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>base_seed</code></strong> : <code>int</code></dt>
|
||
<dd>Random seed.</dd>
|
||
</dl></div>
|
||
</dd>
|
||
</dl>
|
||
</section>
|
||
<section>
|
||
<h2 class="section-title" id="header-classes">Classes</h2>
|
||
<dl>
|
||
<dt id="audiocraft.utils.utils.DummyPoolExecutor"><code class="flex name class">
|
||
<span>class <span class="ident">DummyPoolExecutor</span></span>
|
||
<span>(</span><span>workers, mp_context=None)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">class DummyPoolExecutor:
|
||
"""Dummy pool executor to use when we actually have only 1 worker.
|
||
(e.g. instead of ProcessPoolExecutor).
|
||
"""
|
||
class DummyResult:
|
||
def __init__(self, func, *args, **kwargs):
|
||
self.func = func
|
||
self.args = args
|
||
self.kwargs = kwargs
|
||
|
||
def result(self):
|
||
return self.func(*self.args, **self.kwargs)
|
||
|
||
def __init__(self, workers, mp_context=None):
|
||
pass
|
||
|
||
def submit(self, func, *args, **kwargs):
|
||
return DummyPoolExecutor.DummyResult(func, *args, **kwargs)
|
||
|
||
def __enter__(self):
|
||
return self
|
||
|
||
def __exit__(self, exc_type, exc_value, exc_tb):
|
||
return</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Dummy pool executor to use when we actually have only 1 worker.
|
||
(e.g. instead of ProcessPoolExecutor).</p></div>
|
||
<h3>Class variables</h3>
|
||
<dl>
|
||
<dt id="audiocraft.utils.utils.DummyPoolExecutor.DummyResult"><code class="name">var <span class="ident">DummyResult</span></code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
</dl>
|
||
<h3>Methods</h3>
|
||
<dl>
|
||
<dt id="audiocraft.utils.utils.DummyPoolExecutor.submit"><code class="name flex">
|
||
<span>def <span class="ident">submit</span></span>(<span>self, func, *args, **kwargs)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def submit(self, func, *args, **kwargs):
|
||
return DummyPoolExecutor.DummyResult(func, *args, **kwargs)</code></pre>
|
||
</details>
|
||
<div class="desc"></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.utils" href="index.html">audiocraft.utils</a></code></li>
|
||
</ul>
|
||
</li>
|
||
<li><h3><a href="#header-functions">Functions</a></h3>
|
||
<ul class="">
|
||
<li><code><a title="audiocraft.utils.utils.collate" href="#audiocraft.utils.utils.collate">collate</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.construct_frame_chords" href="#audiocraft.utils.utils.construct_frame_chords">construct_frame_chords</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.copy_state" href="#audiocraft.utils.utils.copy_state">copy_state</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.dict_from_config" href="#audiocraft.utils.utils.dict_from_config">dict_from_config</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.get_dataset_from_loader" href="#audiocraft.utils.utils.get_dataset_from_loader">get_dataset_from_loader</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.get_loader" href="#audiocraft.utils.utils.get_loader">get_loader</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.get_pool_executor" href="#audiocraft.utils.utils.get_pool_executor">get_pool_executor</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.hash_trick" href="#audiocraft.utils.utils.hash_trick">hash_trick</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.is_jsonable" href="#audiocraft.utils.utils.is_jsonable">is_jsonable</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.length_to_mask" href="#audiocraft.utils.utils.length_to_mask">length_to_mask</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.load_clap_state_dict" href="#audiocraft.utils.utils.load_clap_state_dict">load_clap_state_dict</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.model_hash" href="#audiocraft.utils.utils.model_hash">model_hash</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.multinomial" href="#audiocraft.utils.utils.multinomial">multinomial</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.random_subset" href="#audiocraft.utils.utils.random_subset">random_subset</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.sample_top_k" href="#audiocraft.utils.utils.sample_top_k">sample_top_k</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.sample_top_p" href="#audiocraft.utils.utils.sample_top_p">sample_top_p</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.swap_state" href="#audiocraft.utils.utils.swap_state">swap_state</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.warn_once" href="#audiocraft.utils.utils.warn_once">warn_once</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.with_rank_rng" href="#audiocraft.utils.utils.with_rank_rng">with_rank_rng</a></code></li>
|
||
</ul>
|
||
</li>
|
||
<li><h3><a href="#header-classes">Classes</a></h3>
|
||
<ul>
|
||
<li>
|
||
<h4><code><a title="audiocraft.utils.utils.DummyPoolExecutor" href="#audiocraft.utils.utils.DummyPoolExecutor">DummyPoolExecutor</a></code></h4>
|
||
<ul class="">
|
||
<li><code><a title="audiocraft.utils.utils.DummyPoolExecutor.DummyResult" href="#audiocraft.utils.utils.DummyPoolExecutor.DummyResult">DummyResult</a></code></li>
|
||
<li><code><a title="audiocraft.utils.utils.DummyPoolExecutor.submit" href="#audiocraft.utils.utils.DummyPoolExecutor.submit">submit</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>
|