facebookresearch--audiocraft
Alexandre Défossez (autodoc)
680 行
42 KiB
HTML
680 行
42 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.models.musicgen API documentation</title>
|
||
<meta name="description" content="Main model for using MusicGen. This will combine all the required components
|
||
and provide easy access to the generation API.">
|
||
<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.models.musicgen</code></h1>
|
||
</header>
|
||
<section id="section-intro">
|
||
<p>Main model for using MusicGen. This will combine all the required components
|
||
and provide easy access to the generation API.</p>
|
||
</section>
|
||
<section>
|
||
</section>
|
||
<section>
|
||
</section>
|
||
<section>
|
||
</section>
|
||
<section>
|
||
<h2 class="section-title" id="header-classes">Classes</h2>
|
||
<dl>
|
||
<dt id="audiocraft.models.musicgen.MusicGen"><code class="flex name class">
|
||
<span>class <span class="ident">MusicGen</span></span>
|
||
<span>(</span><span>name: str,<br>compression_model: <a title="audiocraft.models.encodec.CompressionModel" href="encodec.html#audiocraft.models.encodec.CompressionModel">CompressionModel</a>,<br>lm: <a title="audiocraft.models.lm.LMModel" href="lm.html#audiocraft.models.lm.LMModel">LMModel</a>,<br>max_duration: float | None = None)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">class MusicGen(BaseGenModel):
|
||
"""MusicGen main model with convenient generation API.
|
||
|
||
Args:
|
||
name (str): name of the model.
|
||
compression_model (CompressionModel): Compression model
|
||
used to map audio to invertible discrete representations.
|
||
lm (LMModel): Language model over discrete representations.
|
||
max_duration (float, optional): maximum duration the model can produce,
|
||
otherwise, inferred from the training params.
|
||
"""
|
||
def __init__(self, name: str, compression_model: CompressionModel, lm: LMModel,
|
||
max_duration: tp.Optional[float] = None):
|
||
super().__init__(name, compression_model, lm, max_duration)
|
||
self.set_generation_params(duration=15) # default duration
|
||
|
||
@staticmethod
|
||
def get_pretrained(name: str = 'facebook/musicgen-melody', device=None):
|
||
"""Return pretrained model, we provide four models:
|
||
- facebook/musicgen-small (300M), text to music,
|
||
# see: https://huggingface.co/facebook/musicgen-small
|
||
- facebook/musicgen-medium (1.5B), text to music,
|
||
# see: https://huggingface.co/facebook/musicgen-medium
|
||
- facebook/musicgen-melody (1.5B) text to music and text+melody to music,
|
||
# see: https://huggingface.co/facebook/musicgen-melody
|
||
- facebook/musicgen-large (3.3B), text to music,
|
||
# see: https://huggingface.co/facebook/musicgen-large
|
||
- facebook/musicgen-style (1.5 B), text and style to music,
|
||
# see: https://huggingface.co/facebook/musicgen-style
|
||
"""
|
||
if device is None:
|
||
if torch.cuda.device_count():
|
||
device = 'cuda'
|
||
else:
|
||
device = 'cpu'
|
||
|
||
if name == 'debug':
|
||
# used only for unit tests
|
||
compression_model = get_debug_compression_model(device)
|
||
lm = get_debug_lm_model(device)
|
||
return MusicGen(name, compression_model, lm, max_duration=30)
|
||
|
||
if name in _HF_MODEL_CHECKPOINTS_MAP:
|
||
warnings.warn(
|
||
"MusicGen pretrained model relying on deprecated checkpoint mapping. " +
|
||
f"Please use full pre-trained id instead: facebook/musicgen-{name}")
|
||
name = _HF_MODEL_CHECKPOINTS_MAP[name]
|
||
|
||
lm = load_lm_model(name, device=device)
|
||
compression_model = load_compression_model(name, device=device)
|
||
if 'self_wav' in lm.condition_provider.conditioners:
|
||
lm.condition_provider.conditioners['self_wav'].match_len_on_eval = True
|
||
lm.condition_provider.conditioners['self_wav']._use_masking = False
|
||
|
||
return MusicGen(name, compression_model, lm)
|
||
|
||
def set_generation_params(self, use_sampling: bool = True, top_k: int = 250,
|
||
top_p: float = 0.0, temperature: float = 1.0,
|
||
duration: float = 30.0, cfg_coef: float = 3.0,
|
||
cfg_coef_beta: tp.Optional[float] = None,
|
||
two_step_cfg: bool = False, extend_stride: float = 18,):
|
||
"""Set the generation parameters for MusicGen.
|
||
|
||
Args:
|
||
use_sampling (bool, optional): Use sampling if True, else do argmax decoding. Defaults to True.
|
||
top_k (int, optional): top_k used for sampling. Defaults to 250.
|
||
top_p (float, optional): top_p used for sampling, when set to 0 top_k is used. Defaults to 0.0.
|
||
temperature (float, optional): Softmax temperature parameter. Defaults to 1.0.
|
||
duration (float, optional): Duration of the generated waveform. Defaults to 30.0.
|
||
cfg_coef (float, optional): Coefficient used for classifier free guidance. Defaults to 3.0.
|
||
cfg_coef_beta (float, optional): beta coefficient in double classifier free guidance.
|
||
Should be only used for MusicGen melody if we want to push the text condition more than
|
||
the audio conditioning. See paragraph 4.3 in https://arxiv.org/pdf/2407.12563 to understand
|
||
double CFG.
|
||
two_step_cfg (bool, optional): If True, performs 2 forward for Classifier Free Guidance,
|
||
instead of batching together the two. This has some impact on how things
|
||
are padded but seems to have little impact in practice.
|
||
extend_stride: when doing extended generation (i.e. more than 30 seconds), by how much
|
||
should we extend the audio each time. Larger values will mean less context is
|
||
preserved, and shorter value will require extra computations.
|
||
"""
|
||
assert extend_stride < self.max_duration, "Cannot stride by more than max generation duration."
|
||
self.extend_stride = extend_stride
|
||
self.duration = duration
|
||
self.generation_params = {
|
||
'use_sampling': use_sampling,
|
||
'temp': temperature,
|
||
'top_k': top_k,
|
||
'top_p': top_p,
|
||
'cfg_coef': cfg_coef,
|
||
'two_step_cfg': two_step_cfg,
|
||
'cfg_coef_beta': cfg_coef_beta,
|
||
}
|
||
|
||
def set_style_conditioner_params(self, eval_q: int = 3, excerpt_length: float = 3.0,
|
||
ds_factor: tp.Optional[int] = None,
|
||
encodec_n_q: tp.Optional[int] = None) -> None:
|
||
"""Set the parameters of the style conditioner
|
||
Args:
|
||
eval_q (int): the number of residual quantization streams used to quantize the style condition
|
||
the smaller it is, the narrower is the information bottleneck
|
||
excerpt_length (float): the excerpt length in seconds that is extracted from the audio
|
||
conditioning
|
||
ds_factor: (int): the downsampling factor used to downsample the style tokens before
|
||
using them as a prefix
|
||
encodec_n_q: (int, optional): if encodec is used as a feature extractor, sets the number
|
||
of streams that is used to extract features
|
||
"""
|
||
assert isinstance(self.lm.condition_provider.conditioners.self_wav, StyleConditioner), \
|
||
"Only use this function if you model is MusicGen-Style"
|
||
self.lm.condition_provider.conditioners.self_wav.set_params(eval_q=eval_q,
|
||
excerpt_length=excerpt_length,
|
||
ds_factor=ds_factor,
|
||
encodec_n_q=encodec_n_q)
|
||
|
||
def generate_with_chroma(self, descriptions: tp.List[str], melody_wavs: MelodyType,
|
||
melody_sample_rate: int, progress: bool = False,
|
||
return_tokens: bool = False) -> tp.Union[torch.Tensor,
|
||
tp.Tuple[torch.Tensor, torch.Tensor]]:
|
||
"""Generate samples conditioned on text and melody.
|
||
|
||
Args:
|
||
descriptions (list of str): A list of strings used as text conditioning.
|
||
melody_wavs: (torch.Tensor or list of Tensor): A batch of waveforms used as
|
||
melody conditioning. Should have shape [B, C, T] with B matching the description length,
|
||
C=1 or 2. It can be [C, T] if there is a single description. It can also be
|
||
a list of [C, T] tensors.
|
||
melody_sample_rate: (int): Sample rate of the melody waveforms.
|
||
progress (bool, optional): Flag to display progress of the generation process. Defaults to False.
|
||
"""
|
||
if isinstance(melody_wavs, torch.Tensor):
|
||
if melody_wavs.dim() == 2:
|
||
melody_wavs = melody_wavs[None]
|
||
if melody_wavs.dim() != 3:
|
||
raise ValueError("Melody wavs should have a shape [B, C, T].")
|
||
melody_wavs = list(melody_wavs)
|
||
else:
|
||
for melody in melody_wavs:
|
||
if melody is not None:
|
||
assert melody.dim() == 2, "One melody in the list has the wrong number of dims."
|
||
|
||
melody_wavs = [
|
||
convert_audio(wav, melody_sample_rate, self.sample_rate, self.audio_channels)
|
||
if wav is not None else None
|
||
for wav in melody_wavs]
|
||
attributes, prompt_tokens = self._prepare_tokens_and_attributes(descriptions=descriptions, prompt=None,
|
||
melody_wavs=melody_wavs)
|
||
assert prompt_tokens is None
|
||
tokens = self._generate_tokens(attributes, prompt_tokens, progress)
|
||
if return_tokens:
|
||
return self.generate_audio(tokens), tokens
|
||
return self.generate_audio(tokens)
|
||
|
||
@torch.no_grad()
|
||
def _prepare_tokens_and_attributes(
|
||
self,
|
||
descriptions: tp.Sequence[tp.Optional[str]],
|
||
prompt: tp.Optional[torch.Tensor],
|
||
melody_wavs: tp.Optional[MelodyList] = None,
|
||
) -> tp.Tuple[tp.List[ConditioningAttributes], tp.Optional[torch.Tensor]]:
|
||
"""Prepare model inputs.
|
||
|
||
Args:
|
||
descriptions (list of str): A list of strings used as text conditioning.
|
||
prompt (torch.Tensor): A batch of waveforms used for continuation.
|
||
melody_wavs (torch.Tensor, optional): A batch of waveforms
|
||
used as melody conditioning. Defaults to None.
|
||
"""
|
||
attributes = [
|
||
ConditioningAttributes(text={'description': description})
|
||
for description in descriptions]
|
||
|
||
if melody_wavs is None:
|
||
for attr in attributes:
|
||
attr.wav['self_wav'] = WavCondition(
|
||
torch.zeros((1, 1, 1), device=self.device),
|
||
torch.tensor([0], device=self.device),
|
||
sample_rate=[self.sample_rate],
|
||
path=[None])
|
||
else:
|
||
if 'self_wav' not in self.lm.condition_provider.conditioners:
|
||
raise RuntimeError("This model doesn't support melody conditioning. "
|
||
"Use the `melody` model.")
|
||
assert len(melody_wavs) == len(descriptions), \
|
||
f"number of melody wavs must match number of descriptions! " \
|
||
f"got melody len={len(melody_wavs)}, and descriptions len={len(descriptions)}"
|
||
for attr, melody in zip(attributes, melody_wavs):
|
||
if melody is None:
|
||
attr.wav['self_wav'] = WavCondition(
|
||
torch.zeros((1, 1, 1), device=self.device),
|
||
torch.tensor([0], device=self.device),
|
||
sample_rate=[self.sample_rate],
|
||
path=[None])
|
||
else:
|
||
attr.wav['self_wav'] = WavCondition(
|
||
melody[None].to(device=self.device),
|
||
torch.tensor([melody.shape[-1]], device=self.device),
|
||
sample_rate=[self.sample_rate],
|
||
path=[None],
|
||
)
|
||
|
||
if prompt is not None:
|
||
if descriptions is not None:
|
||
assert len(descriptions) == len(prompt), "Prompt and nb. descriptions doesn't match"
|
||
prompt = prompt.to(self.device)
|
||
prompt_tokens, scale = self.compression_model.encode(prompt)
|
||
assert scale is None
|
||
else:
|
||
prompt_tokens = None
|
||
return attributes, prompt_tokens
|
||
|
||
def _generate_tokens(self, attributes: tp.List[ConditioningAttributes],
|
||
prompt_tokens: tp.Optional[torch.Tensor], progress: bool = False) -> torch.Tensor:
|
||
"""Generate discrete audio tokens given audio prompt and/or conditions.
|
||
|
||
Args:
|
||
attributes (list of ConditioningAttributes): Conditions used for generation (text/melody).
|
||
prompt_tokens (torch.Tensor, optional): Audio prompt used for continuation.
|
||
progress (bool, optional): Flag to display progress of the generation process. Defaults to False.
|
||
Returns:
|
||
torch.Tensor: Generated audio, of shape [B, C, T], T is defined by the generation params.
|
||
"""
|
||
total_gen_len = int(self.duration * self.frame_rate)
|
||
max_prompt_len = int(min(self.duration, self.max_duration) * self.frame_rate)
|
||
current_gen_offset: int = 0
|
||
|
||
def _progress_callback(generated_tokens: int, tokens_to_generate: int):
|
||
generated_tokens += current_gen_offset
|
||
if self._progress_callback is not None:
|
||
# Note that total_gen_len might be quite wrong depending on the
|
||
# codebook pattern used, but with delay it is almost accurate.
|
||
self._progress_callback(generated_tokens, tokens_to_generate)
|
||
else:
|
||
print(f'{generated_tokens: 6d} / {tokens_to_generate: 6d}', end='\r')
|
||
|
||
if prompt_tokens is not None:
|
||
assert max_prompt_len >= prompt_tokens.shape[-1], \
|
||
"Prompt is longer than audio to generate"
|
||
|
||
callback = None
|
||
if progress:
|
||
callback = _progress_callback
|
||
|
||
if self.duration <= self.max_duration:
|
||
# generate by sampling from LM, simple case.
|
||
with self.autocast:
|
||
gen_tokens = self.lm.generate(
|
||
prompt_tokens, attributes,
|
||
callback=callback, max_gen_len=total_gen_len, **self.generation_params)
|
||
|
||
else:
|
||
# now this gets a bit messier, we need to handle prompts,
|
||
# melody conditioning etc.
|
||
ref_wavs = [attr.wav['self_wav'] for attr in attributes]
|
||
all_tokens = []
|
||
if prompt_tokens is None:
|
||
prompt_length = 0
|
||
else:
|
||
all_tokens.append(prompt_tokens)
|
||
prompt_length = prompt_tokens.shape[-1]
|
||
|
||
assert self.extend_stride is not None, "Stride should be defined to generate beyond max_duration"
|
||
assert self.extend_stride < self.max_duration, "Cannot stride by more than max generation duration."
|
||
stride_tokens = int(self.frame_rate * self.extend_stride)
|
||
|
||
while current_gen_offset + prompt_length < total_gen_len:
|
||
time_offset = current_gen_offset / self.frame_rate
|
||
chunk_duration = min(self.duration - time_offset, self.max_duration)
|
||
max_gen_len = int(chunk_duration * self.frame_rate)
|
||
for attr, ref_wav in zip(attributes, ref_wavs):
|
||
wav_length = ref_wav.length.item()
|
||
if wav_length == 0:
|
||
continue
|
||
# We will extend the wav periodically if it not long enough.
|
||
# we have to do it here rather than in conditioners.py as otherwise
|
||
# we wouldn't have the full wav.
|
||
initial_position = int(time_offset * self.sample_rate)
|
||
wav_target_length = int(self.max_duration * self.sample_rate)
|
||
positions = torch.arange(initial_position,
|
||
initial_position + wav_target_length, device=self.device)
|
||
attr.wav['self_wav'] = WavCondition(
|
||
ref_wav[0][..., positions % wav_length],
|
||
torch.full_like(ref_wav[1], wav_target_length),
|
||
[self.sample_rate] * ref_wav[0].size(0),
|
||
[None], [0.])
|
||
with self.autocast:
|
||
gen_tokens = self.lm.generate(
|
||
prompt_tokens, attributes,
|
||
callback=callback, max_gen_len=max_gen_len, **self.generation_params)
|
||
if prompt_tokens is None:
|
||
all_tokens.append(gen_tokens)
|
||
else:
|
||
all_tokens.append(gen_tokens[:, :, prompt_tokens.shape[-1]:])
|
||
prompt_tokens = gen_tokens[:, :, stride_tokens:]
|
||
prompt_length = prompt_tokens.shape[-1]
|
||
current_gen_offset += stride_tokens
|
||
|
||
gen_tokens = torch.cat(all_tokens, dim=-1)
|
||
return gen_tokens</code></pre>
|
||
</details>
|
||
<div class="desc"><p>MusicGen main model with convenient generation API.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>name</code></strong> : <code>str</code></dt>
|
||
<dd>name of the model.</dd>
|
||
<dt><strong><code>compression_model</code></strong> : <code>CompressionModel</code></dt>
|
||
<dd>Compression model
|
||
used to map audio to invertible discrete representations.</dd>
|
||
<dt><strong><code>lm</code></strong> : <code>LMModel</code></dt>
|
||
<dd>Language model over discrete representations.</dd>
|
||
<dt><strong><code>max_duration</code></strong> : <code>float</code>, optional</dt>
|
||
<dd>maximum duration the model can produce,
|
||
otherwise, inferred from the training params.</dd>
|
||
</dl></div>
|
||
<h3>Ancestors</h3>
|
||
<ul class="hlist">
|
||
<li><a title="audiocraft.models.genmodel.BaseGenModel" href="genmodel.html#audiocraft.models.genmodel.BaseGenModel">BaseGenModel</a></li>
|
||
<li>abc.ABC</li>
|
||
</ul>
|
||
<h3>Static methods</h3>
|
||
<dl>
|
||
<dt id="audiocraft.models.musicgen.MusicGen.get_pretrained"><code class="name flex">
|
||
<span>def <span class="ident">get_pretrained</span></span>(<span>name: str = 'facebook/musicgen-melody', device=None)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">@staticmethod
|
||
def get_pretrained(name: str = 'facebook/musicgen-melody', device=None):
|
||
"""Return pretrained model, we provide four models:
|
||
- facebook/musicgen-small (300M), text to music,
|
||
# see: https://huggingface.co/facebook/musicgen-small
|
||
- facebook/musicgen-medium (1.5B), text to music,
|
||
# see: https://huggingface.co/facebook/musicgen-medium
|
||
- facebook/musicgen-melody (1.5B) text to music and text+melody to music,
|
||
# see: https://huggingface.co/facebook/musicgen-melody
|
||
- facebook/musicgen-large (3.3B), text to music,
|
||
# see: https://huggingface.co/facebook/musicgen-large
|
||
- facebook/musicgen-style (1.5 B), text and style to music,
|
||
# see: https://huggingface.co/facebook/musicgen-style
|
||
"""
|
||
if device is None:
|
||
if torch.cuda.device_count():
|
||
device = 'cuda'
|
||
else:
|
||
device = 'cpu'
|
||
|
||
if name == 'debug':
|
||
# used only for unit tests
|
||
compression_model = get_debug_compression_model(device)
|
||
lm = get_debug_lm_model(device)
|
||
return MusicGen(name, compression_model, lm, max_duration=30)
|
||
|
||
if name in _HF_MODEL_CHECKPOINTS_MAP:
|
||
warnings.warn(
|
||
"MusicGen pretrained model relying on deprecated checkpoint mapping. " +
|
||
f"Please use full pre-trained id instead: facebook/musicgen-{name}")
|
||
name = _HF_MODEL_CHECKPOINTS_MAP[name]
|
||
|
||
lm = load_lm_model(name, device=device)
|
||
compression_model = load_compression_model(name, device=device)
|
||
if 'self_wav' in lm.condition_provider.conditioners:
|
||
lm.condition_provider.conditioners['self_wav'].match_len_on_eval = True
|
||
lm.condition_provider.conditioners['self_wav']._use_masking = False
|
||
|
||
return MusicGen(name, compression_model, lm)</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Return pretrained model, we provide four models:
|
||
- facebook/musicgen-small (300M), text to music,
|
||
# see: <a href="https://huggingface.co/facebook/musicgen-small">https://huggingface.co/facebook/musicgen-small</a>
|
||
- facebook/musicgen-medium (1.5B), text to music,
|
||
# see: <a href="https://huggingface.co/facebook/musicgen-medium">https://huggingface.co/facebook/musicgen-medium</a>
|
||
- facebook/musicgen-melody (1.5B) text to music and text+melody to music,
|
||
# see: <a href="https://huggingface.co/facebook/musicgen-melody">https://huggingface.co/facebook/musicgen-melody</a>
|
||
- facebook/musicgen-large (3.3B), text to music,
|
||
# see: <a href="https://huggingface.co/facebook/musicgen-large">https://huggingface.co/facebook/musicgen-large</a>
|
||
- facebook/musicgen-style (1.5 B), text and style to music,
|
||
# see: <a href="https://huggingface.co/facebook/musicgen-style">https://huggingface.co/facebook/musicgen-style</a></p></div>
|
||
</dd>
|
||
</dl>
|
||
<h3>Methods</h3>
|
||
<dl>
|
||
<dt id="audiocraft.models.musicgen.MusicGen.generate_with_chroma"><code class="name flex">
|
||
<span>def <span class="ident">generate_with_chroma</span></span>(<span>self,<br>descriptions: List[str],<br>melody_wavs: torch.Tensor | List[torch.Tensor | None],<br>melody_sample_rate: int,<br>progress: bool = False,<br>return_tokens: bool = False) ‑> torch.Tensor | Tuple[torch.Tensor, torch.Tensor]</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def generate_with_chroma(self, descriptions: tp.List[str], melody_wavs: MelodyType,
|
||
melody_sample_rate: int, progress: bool = False,
|
||
return_tokens: bool = False) -> tp.Union[torch.Tensor,
|
||
tp.Tuple[torch.Tensor, torch.Tensor]]:
|
||
"""Generate samples conditioned on text and melody.
|
||
|
||
Args:
|
||
descriptions (list of str): A list of strings used as text conditioning.
|
||
melody_wavs: (torch.Tensor or list of Tensor): A batch of waveforms used as
|
||
melody conditioning. Should have shape [B, C, T] with B matching the description length,
|
||
C=1 or 2. It can be [C, T] if there is a single description. It can also be
|
||
a list of [C, T] tensors.
|
||
melody_sample_rate: (int): Sample rate of the melody waveforms.
|
||
progress (bool, optional): Flag to display progress of the generation process. Defaults to False.
|
||
"""
|
||
if isinstance(melody_wavs, torch.Tensor):
|
||
if melody_wavs.dim() == 2:
|
||
melody_wavs = melody_wavs[None]
|
||
if melody_wavs.dim() != 3:
|
||
raise ValueError("Melody wavs should have a shape [B, C, T].")
|
||
melody_wavs = list(melody_wavs)
|
||
else:
|
||
for melody in melody_wavs:
|
||
if melody is not None:
|
||
assert melody.dim() == 2, "One melody in the list has the wrong number of dims."
|
||
|
||
melody_wavs = [
|
||
convert_audio(wav, melody_sample_rate, self.sample_rate, self.audio_channels)
|
||
if wav is not None else None
|
||
for wav in melody_wavs]
|
||
attributes, prompt_tokens = self._prepare_tokens_and_attributes(descriptions=descriptions, prompt=None,
|
||
melody_wavs=melody_wavs)
|
||
assert prompt_tokens is None
|
||
tokens = self._generate_tokens(attributes, prompt_tokens, progress)
|
||
if return_tokens:
|
||
return self.generate_audio(tokens), tokens
|
||
return self.generate_audio(tokens)</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Generate samples conditioned on text and melody.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>descriptions</code></strong> : <code>list</code> of <code>str</code></dt>
|
||
<dd>A list of strings used as text conditioning.</dd>
|
||
<dt><strong><code>melody_wavs</code></strong></dt>
|
||
<dd>(torch.Tensor or list of Tensor): A batch of waveforms used as
|
||
melody conditioning. Should have shape [B, C, T] with B matching the description length,
|
||
C=1 or 2. It can be [C, T] if there is a single description. It can also be
|
||
a list of [C, T] tensors.</dd>
|
||
<dt><strong><code>melody_sample_rate</code></strong></dt>
|
||
<dd>(int): Sample rate of the melody waveforms.</dd>
|
||
<dt><strong><code>progress</code></strong> : <code>bool</code>, optional</dt>
|
||
<dd>Flag to display progress of the generation process. Defaults to False.</dd>
|
||
</dl></div>
|
||
</dd>
|
||
<dt id="audiocraft.models.musicgen.MusicGen.set_generation_params"><code class="name flex">
|
||
<span>def <span class="ident">set_generation_params</span></span>(<span>self,<br>use_sampling: bool = True,<br>top_k: int = 250,<br>top_p: float = 0.0,<br>temperature: float = 1.0,<br>duration: float = 30.0,<br>cfg_coef: float = 3.0,<br>cfg_coef_beta: float | None = None,<br>two_step_cfg: bool = False,<br>extend_stride: float = 18)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def set_generation_params(self, use_sampling: bool = True, top_k: int = 250,
|
||
top_p: float = 0.0, temperature: float = 1.0,
|
||
duration: float = 30.0, cfg_coef: float = 3.0,
|
||
cfg_coef_beta: tp.Optional[float] = None,
|
||
two_step_cfg: bool = False, extend_stride: float = 18,):
|
||
"""Set the generation parameters for MusicGen.
|
||
|
||
Args:
|
||
use_sampling (bool, optional): Use sampling if True, else do argmax decoding. Defaults to True.
|
||
top_k (int, optional): top_k used for sampling. Defaults to 250.
|
||
top_p (float, optional): top_p used for sampling, when set to 0 top_k is used. Defaults to 0.0.
|
||
temperature (float, optional): Softmax temperature parameter. Defaults to 1.0.
|
||
duration (float, optional): Duration of the generated waveform. Defaults to 30.0.
|
||
cfg_coef (float, optional): Coefficient used for classifier free guidance. Defaults to 3.0.
|
||
cfg_coef_beta (float, optional): beta coefficient in double classifier free guidance.
|
||
Should be only used for MusicGen melody if we want to push the text condition more than
|
||
the audio conditioning. See paragraph 4.3 in https://arxiv.org/pdf/2407.12563 to understand
|
||
double CFG.
|
||
two_step_cfg (bool, optional): If True, performs 2 forward for Classifier Free Guidance,
|
||
instead of batching together the two. This has some impact on how things
|
||
are padded but seems to have little impact in practice.
|
||
extend_stride: when doing extended generation (i.e. more than 30 seconds), by how much
|
||
should we extend the audio each time. Larger values will mean less context is
|
||
preserved, and shorter value will require extra computations.
|
||
"""
|
||
assert extend_stride < self.max_duration, "Cannot stride by more than max generation duration."
|
||
self.extend_stride = extend_stride
|
||
self.duration = duration
|
||
self.generation_params = {
|
||
'use_sampling': use_sampling,
|
||
'temp': temperature,
|
||
'top_k': top_k,
|
||
'top_p': top_p,
|
||
'cfg_coef': cfg_coef,
|
||
'two_step_cfg': two_step_cfg,
|
||
'cfg_coef_beta': cfg_coef_beta,
|
||
}</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Set the generation parameters for MusicGen.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>use_sampling</code></strong> : <code>bool</code>, optional</dt>
|
||
<dd>Use sampling if True, else do argmax decoding. Defaults to True.</dd>
|
||
<dt><strong><code>top_k</code></strong> : <code>int</code>, optional</dt>
|
||
<dd>top_k used for sampling. Defaults to 250.</dd>
|
||
<dt><strong><code>top_p</code></strong> : <code>float</code>, optional</dt>
|
||
<dd>top_p used for sampling, when set to 0 top_k is used. Defaults to 0.0.</dd>
|
||
<dt><strong><code>temperature</code></strong> : <code>float</code>, optional</dt>
|
||
<dd>Softmax temperature parameter. Defaults to 1.0.</dd>
|
||
<dt><strong><code>duration</code></strong> : <code>float</code>, optional</dt>
|
||
<dd>Duration of the generated waveform. Defaults to 30.0.</dd>
|
||
<dt><strong><code>cfg_coef</code></strong> : <code>float</code>, optional</dt>
|
||
<dd>Coefficient used for classifier free guidance. Defaults to 3.0.</dd>
|
||
<dt><strong><code>cfg_coef_beta</code></strong> : <code>float</code>, optional</dt>
|
||
<dd>beta coefficient in double classifier free guidance.
|
||
Should be only used for MusicGen melody if we want to push the text condition more than
|
||
the audio conditioning. See paragraph 4.3 in <a href="https://arxiv.org/pdf/2407.12563">https://arxiv.org/pdf/2407.12563</a> to understand
|
||
double CFG.</dd>
|
||
<dt><strong><code>two_step_cfg</code></strong> : <code>bool</code>, optional</dt>
|
||
<dd>If True, performs 2 forward for Classifier Free Guidance,
|
||
instead of batching together the two. This has some impact on how things
|
||
are padded but seems to have little impact in practice.</dd>
|
||
<dt><strong><code>extend_stride</code></strong></dt>
|
||
<dd>when doing extended generation (i.e. more than 30 seconds), by how much
|
||
should we extend the audio each time. Larger values will mean less context is
|
||
preserved, and shorter value will require extra computations.</dd>
|
||
</dl></div>
|
||
</dd>
|
||
<dt id="audiocraft.models.musicgen.MusicGen.set_style_conditioner_params"><code class="name flex">
|
||
<span>def <span class="ident">set_style_conditioner_params</span></span>(<span>self,<br>eval_q: int = 3,<br>excerpt_length: float = 3.0,<br>ds_factor: int | None = None,<br>encodec_n_q: int | None = None) ‑> None</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def set_style_conditioner_params(self, eval_q: int = 3, excerpt_length: float = 3.0,
|
||
ds_factor: tp.Optional[int] = None,
|
||
encodec_n_q: tp.Optional[int] = None) -> None:
|
||
"""Set the parameters of the style conditioner
|
||
Args:
|
||
eval_q (int): the number of residual quantization streams used to quantize the style condition
|
||
the smaller it is, the narrower is the information bottleneck
|
||
excerpt_length (float): the excerpt length in seconds that is extracted from the audio
|
||
conditioning
|
||
ds_factor: (int): the downsampling factor used to downsample the style tokens before
|
||
using them as a prefix
|
||
encodec_n_q: (int, optional): if encodec is used as a feature extractor, sets the number
|
||
of streams that is used to extract features
|
||
"""
|
||
assert isinstance(self.lm.condition_provider.conditioners.self_wav, StyleConditioner), \
|
||
"Only use this function if you model is MusicGen-Style"
|
||
self.lm.condition_provider.conditioners.self_wav.set_params(eval_q=eval_q,
|
||
excerpt_length=excerpt_length,
|
||
ds_factor=ds_factor,
|
||
encodec_n_q=encodec_n_q)</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Set the parameters of the style conditioner</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>eval_q</code></strong> : <code>int</code></dt>
|
||
<dd>the number of residual quantization streams used to quantize the style condition
|
||
the smaller it is, the narrower is the information bottleneck</dd>
|
||
<dt><strong><code>excerpt_length</code></strong> : <code>float</code></dt>
|
||
<dd>the excerpt length in seconds that is extracted from the audio
|
||
conditioning</dd>
|
||
<dt><strong><code>ds_factor</code></strong></dt>
|
||
<dd>(int): the downsampling factor used to downsample the style tokens before
|
||
using them as a prefix</dd>
|
||
<dt><strong><code>encodec_n_q</code></strong></dt>
|
||
<dd>(int, optional): if encodec is used as a feature extractor, sets the number
|
||
of streams that is used to extract features</dd>
|
||
</dl></div>
|
||
</dd>
|
||
</dl>
|
||
<h3>Inherited members</h3>
|
||
<ul class="hlist">
|
||
<li><code><b><a title="audiocraft.models.genmodel.BaseGenModel" href="genmodel.html#audiocraft.models.genmodel.BaseGenModel">BaseGenModel</a></b></code>:
|
||
<ul class="hlist">
|
||
<li><code><a title="audiocraft.models.genmodel.BaseGenModel.audio_channels" href="genmodel.html#audiocraft.models.genmodel.BaseGenModel.audio_channels">audio_channels</a></code></li>
|
||
<li><code><a title="audiocraft.models.genmodel.BaseGenModel.frame_rate" href="genmodel.html#audiocraft.models.genmodel.BaseGenModel.frame_rate">frame_rate</a></code></li>
|
||
<li><code><a title="audiocraft.models.genmodel.BaseGenModel.generate" href="genmodel.html#audiocraft.models.genmodel.BaseGenModel.generate">generate</a></code></li>
|
||
<li><code><a title="audiocraft.models.genmodel.BaseGenModel.generate_audio" href="genmodel.html#audiocraft.models.genmodel.BaseGenModel.generate_audio">generate_audio</a></code></li>
|
||
<li><code><a title="audiocraft.models.genmodel.BaseGenModel.generate_continuation" href="genmodel.html#audiocraft.models.genmodel.BaseGenModel.generate_continuation">generate_continuation</a></code></li>
|
||
<li><code><a title="audiocraft.models.genmodel.BaseGenModel.generate_unconditional" href="genmodel.html#audiocraft.models.genmodel.BaseGenModel.generate_unconditional">generate_unconditional</a></code></li>
|
||
<li><code><a title="audiocraft.models.genmodel.BaseGenModel.sample_rate" href="genmodel.html#audiocraft.models.genmodel.BaseGenModel.sample_rate">sample_rate</a></code></li>
|
||
<li><code><a title="audiocraft.models.genmodel.BaseGenModel.set_custom_progress_callback" href="genmodel.html#audiocraft.models.genmodel.BaseGenModel.set_custom_progress_callback">set_custom_progress_callback</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.models" href="index.html">audiocraft.models</a></code></li>
|
||
</ul>
|
||
</li>
|
||
<li><h3><a href="#header-classes">Classes</a></h3>
|
||
<ul>
|
||
<li>
|
||
<h4><code><a title="audiocraft.models.musicgen.MusicGen" href="#audiocraft.models.musicgen.MusicGen">MusicGen</a></code></h4>
|
||
<ul class="">
|
||
<li><code><a title="audiocraft.models.musicgen.MusicGen.generate_with_chroma" href="#audiocraft.models.musicgen.MusicGen.generate_with_chroma">generate_with_chroma</a></code></li>
|
||
<li><code><a title="audiocraft.models.musicgen.MusicGen.get_pretrained" href="#audiocraft.models.musicgen.MusicGen.get_pretrained">get_pretrained</a></code></li>
|
||
<li><code><a title="audiocraft.models.musicgen.MusicGen.set_generation_params" href="#audiocraft.models.musicgen.MusicGen.set_generation_params">set_generation_params</a></code></li>
|
||
<li><code><a title="audiocraft.models.musicgen.MusicGen.set_style_conditioner_params" href="#audiocraft.models.musicgen.MusicGen.set_style_conditioner_params">set_style_conditioner_params</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>
|