The MusicGen model was proposed in the paper Simple and Controllable Music Generation by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
MusicGen is a single stage auto-regressive Transformer model capable of generating high-quality music samples conditioned on text descriptions or audio prompts. The text descriptions are passed through a frozen text encoder model to obtain a sequence of hidden-state representations. MusicGen is then trained to predict discrete audio tokens, or audio codes, conditioned on these hidden-states. These audio tokens are then decoded using an audio compression model, such as EnCodec, to recover the audio waveform.
Through an efficient token interleaving pattern, MusicGen does not require a self-supervised semantic representation of the text/audio prompts, thus eliminating the need to cascade multiple models to predict a set of codebooks (e.g. hierarchically or upsampling). Instead, it is able to generate all the codebooks in a single forward pass.
The abstract from the paper is the following:
We tackle the task of conditional music generation. We introduce MusicGen, a single Language Model (LM) that operates over several streams of compressed discrete music representation, i.e., tokens. Unlike prior work, MusicGen is comprised of a single-stage transformer LM together with efficient token interleaving patterns, which eliminates the need for cascading several models, e.g., hierarchically or upsampling. Following this approach, we demonstrate how MusicGen can generate high-quality samples, while being conditioned on textual description or melodic features, allowing better controls over the generated output. We conduct extensive empirical evaluation, considering both automatic and human studies, showing the proposed approach is superior to the evaluated baselines on a standard text-to-music benchmark. Through ablation studies, we shed light over the importance of each of the components comprising MusicGen.
This model was contributed by sanchit-gandhi. The original code can be found here. The pre-trained checkpoints can be found on the Hugging Face Hub.
src/transformers/models/musicgen/convert_musicgen_transformers.py with the following command:python src/transformers/models/musicgen/convert_musicgen_transformers.py \
--checkpoint small --pytorch_dump_folder /output/path --safe_serialization MusicGen is compatible with two generation modes: greedy and sampling. In practice, sampling leads to significantly
better results than greedy, thus we encourage sampling mode to be used where possible. Sampling is enabled by default,
and can be explicitly specified by setting do_sample=True in the call to MusicgenForConditionalGeneration.generate(),
or by overriding the model’s generation config (see below).
Generation is limited by the sinusoidal positional embeddings to 30 second inputs. Meaning, MusicGen cannot generate more than 30 seconds of audio (1503 tokens), and input audio passed by Audio-Prompted Generation contributes to this limit so, given an input of 20 seconds of audio, MusicGen cannot generate more than 10 seconds of additional audio.
Transformers supports both mono (1-channel) and stereo (2-channel) variants of MusicGen. The mono channel versions generate a single set of codebooks. The stereo versions generate 2 sets of codebooks, 1 for each channel (left/right), and each set of codebooks is decoded independently through the audio compression model. The audio streams for each channel are combined to give the final stereo output.
The inputs for unconditional (or ‘null’) generation can be obtained through the method
MusicgenForConditionalGeneration.get_unconditional_inputs():
>>> from transformers import MusicgenForConditionalGeneration
>>> model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
>>> unconditional_inputs = model.get_unconditional_inputs(num_samples=1)
>>> audio_values = model.generate(**unconditional_inputs, do_sample=True, max_new_tokens=256)The audio outputs are a three-dimensional Torch tensor of shape (batch_size, num_channels, sequence_length). To listen
to the generated audio samples, you can either play them in an ipynb notebook:
from IPython.display import Audio
sampling_rate = model.config.audio_encoder.sampling_rate
Audio(audio_values[0].numpy(), rate=sampling_rate)Or save them as a .wav file using a third-party library, e.g. scipy:
>>> import scipy
>>> sampling_rate = model.config.audio_encoder.sampling_rate
>>> scipy.io.wavfile.write("musicgen_out.wav", rate=sampling_rate, data=audio_values[0, 0].numpy())The model can generate an audio sample conditioned on a text prompt through use of the MusicgenProcessor to pre-process the inputs:
>>> from transformers import AutoProcessor, MusicgenForConditionalGeneration
>>> processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
>>> model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
>>> inputs = processor(
... text=["80s pop track with bassy drums and synth", "90s rock song with loud guitars and heavy drums"],
... padding=True,
... return_tensors="pt",
... )
>>> audio_values = model.generate(**inputs, do_sample=True, guidance_scale=3, max_new_tokens=256)The guidance_scale is used in classifier free guidance (CFG), setting the weighting between the conditional logits
(which are predicted from the text prompts) and the unconditional logits (which are predicted from an unconditional or
‘null’ prompt). Higher guidance scale encourages the model to generate samples that are more closely linked to the input
prompt, usually at the expense of poorer audio quality. CFG is enabled by setting guidance_scale > 1. For best results,
use guidance_scale=3 (default).
The same MusicgenProcessor can be used to pre-process an audio prompt that is used for audio continuation. In the following example, we load an audio file using the 🤗 Datasets library, which can be pip installed through the command below:
pip install --upgrade pip pip install datasets[audio]
>>> from transformers import AutoProcessor, MusicgenForConditionalGeneration
>>> from datasets import load_dataset
>>> processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
>>> model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
>>> dataset = load_dataset("sanchit-gandhi/gtzan", split="train", streaming=True)
>>> sample = next(iter(dataset))["audio"]
>>> # take the first half of the audio sample
>>> sample["array"] = sample["array"][: len(sample["array"]) // 2]
>>> inputs = processor(
... audio=sample["array"],
... sampling_rate=sample["sampling_rate"],
... text=["80s blues track with groovy saxophone"],
... padding=True,
... return_tensors="pt",
... )
>>> audio_values = model.generate(**inputs, do_sample=True, guidance_scale=3, max_new_tokens=256)For batched audio-prompted generation, the generated audio_values can be post-processed to remove padding by using the
MusicgenProcessor class:
>>> from transformers import AutoProcessor, MusicgenForConditionalGeneration
>>> from datasets import load_dataset
>>> processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
>>> model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
>>> dataset = load_dataset("sanchit-gandhi/gtzan", split="train", streaming=True)
>>> sample = next(iter(dataset))["audio"]
>>> # take the first quarter of the audio sample
>>> sample_1 = sample["array"][: len(sample["array"]) // 4]
>>> # take the first half of the audio sample
>>> sample_2 = sample["array"][: len(sample["array"]) // 2]
>>> inputs = processor(
... audio=[sample_1, sample_2],
... sampling_rate=sample["sampling_rate"],
... text=["80s blues track with groovy saxophone", "90s rock song with loud guitars and heavy drums"],
... padding=True,
... return_tensors="pt",
... )
>>> audio_values = model.generate(**inputs, do_sample=True, guidance_scale=3, max_new_tokens=256)
>>> # post-process to remove padding from the batched audio
>>> audio_values = processor.batch_decode(audio_values, padding_mask=inputs.padding_mask)The default parameters that control the generation process, such as sampling, guidance scale and number of generated tokens, can be found in the model’s generation config, and updated as desired:
>>> from transformers import MusicgenForConditionalGeneration
>>> model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
>>> # inspect the default generation config
>>> model.generation_config
>>> # increase the guidance scale to 4.0
>>> model.generation_config.guidance_scale = 4.0
>>> # decrease the max length to 256 tokens
>>> model.generation_config.max_length = 256Note that any arguments passed to the generate method will supersede those in the generation config, so setting
do_sample=False in the call to generate will supersede the setting of model.generation_config.do_sample in the
generation config.
The MusicGen model can be de-composed into three distinct stages:
Thus, the MusicGen model can either be used as a standalone decoder model, corresponding to the class MusicgenForCausalLM,
or as a composite model that includes the text encoder and audio encoder/decoder, corresponding to the class
MusicgenForConditionalGeneration. If only the decoder needs to be loaded from the pre-trained checkpoint, it can be loaded by first
specifying the correct config, or be accessed through the .decoder attribute of the composite model:
>>> from transformers import AutoConfig, MusicgenForCausalLM, MusicgenForConditionalGeneration
>>> # Option 1: get decoder config and pass to `.from_pretrained`
>>> decoder_config = AutoConfig.from_pretrained("facebook/musicgen-small").decoder
>>> decoder = MusicgenForCausalLM.from_pretrained("facebook/musicgen-small", **decoder_config)
>>> # Option 2: load the entire composite model, but only return the decoder
>>> decoder = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small").decoderSince the text encoder and audio encoder/decoder models are frozen during training, the MusicGen decoder MusicgenForCausalLM can be trained standalone on a dataset of encoder hidden-states and audio codes. For inference, the trained decoder can be combined with the frozen text encoder and audio encoder/decoders to recover the composite MusicgenForConditionalGeneration model.
Tips:
do_sample in the call to MusicgenForConditionalGeneration.generate()( vocab_size = 2048 max_position_embeddings = 2048 num_hidden_layers = 24 ffn_dim = 4096 num_attention_heads = 16 layerdrop = 0.0 use_cache = True activation_function = 'gelu' hidden_size = 1024 dropout = 0.1 attention_dropout = 0.0 activation_dropout = 0.0 initializer_factor = 0.02 scale_embedding = False num_codebooks = 4 audio_channels = 1 pad_token_id = 2048 bos_token_id = 2048 eos_token_id = None tie_word_embeddings = False **kwargs )
Parameters
int, optional, defaults to 2048) —
Vocabulary size of the MusicgenDecoder model. Defines the number of different tokens that can be
represented by the inputs_ids passed when calling MusicgenDecoder. int, optional, defaults to 1024) —
Dimensionality of the layers and the pooler layer. int, optional, defaults to 24) —
Number of decoder layers. int, optional, defaults to 16) —
Number of attention heads for each attention layer in the Transformer block. int, optional, defaults to 4096) —
Dimensionality of the “intermediate” (often named feed-forward) layer in the Transformer block. str or function, optional, defaults to "gelu") —
The non-linear activation function (function or string) in the decoder and pooler. If string, "gelu",
"relu", "silu" and "gelu_new" are supported. float, optional, defaults to 0.1) —
The dropout probability for all fully connected layers in the embeddings, text_encoder, and pooler. float, optional, defaults to 0.0) —
The dropout ratio for the attention probabilities. float, optional, defaults to 0.0) —
The dropout ratio for activations inside the fully connected layer. int, optional, defaults to 2048) —
The maximum sequence length that this model might ever be used with. Typically, set this to something large
just in case (e.g., 512 or 1024 or 2048). float, optional, defaults to 0.02) —
The standard deviation of the truncated_normal_initializer for initializing all weight matrices. float, optional, defaults to 0.0) —
The LayerDrop probability for the decoder. See the [LayerDrop paper](see https://arxiv.org/abs/1909.11556)
for more details. bool, optional, defaults to False) —
Scale embeddings by diving by sqrt(hidden_size). bool, optional, defaults to True) —
Whether the model should return the last key/values attentions (not used by all models) int, optional, defaults to 4) —
The number of parallel codebooks forwarded to the model. bool, optional, defaults to False) —
Whether input and output word embeddings should be tied. int, optional, defaults to 1 —
Number of channels in the audio data. Either 1 for mono or 2 for stereo. Stereo models generate a separate
audio stream for the left/right output channels. Mono models generate a single audio stream output. This is the configuration class to store the configuration of an MusicgenDecoder. It is used to instantiate a
MusicGen decoder according to the specified arguments, defining the model architecture. Instantiating a
configuration with the defaults will yield a similar configuration to that of the MusicGen
facebook/musicgen-small architecture.
Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.
( **kwargs )
Parameters
This is the configuration class to store the configuration of a MusicgenModel. It is used to instantiate a MusicGen model according to the specified arguments, defining the text encoder, audio encoder and MusicGen decoder configs.
Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.
Example:
>>> from transformers import (
... MusicgenConfig,
... MusicgenDecoderConfig,
... T5Config,
... EncodecConfig,
... MusicgenForConditionalGeneration,
... )
>>> # Initializing text encoder, audio encoder, and decoder model configurations
>>> text_encoder_config = T5Config()
>>> audio_encoder_config = EncodecConfig()
>>> decoder_config = MusicgenDecoderConfig()
>>> configuration = MusicgenConfig.from_sub_models_config(
... text_encoder_config, audio_encoder_config, decoder_config
... )
>>> # Initializing a MusicgenForConditionalGeneration (with random weights) from the facebook/musicgen-small style configuration
>>> model = MusicgenForConditionalGeneration(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
>>> config_text_encoder = model.config.text_encoder
>>> config_audio_encoder = model.config.audio_encoder
>>> config_decoder = model.config.decoder
>>> # Saving the model, including its configuration
>>> model.save_pretrained("musicgen-model")
>>> # loading model and config from pretrained folder
>>> musicgen_config = MusicgenConfig.from_pretrained("musicgen-model")
>>> model = MusicgenForConditionalGeneration.from_pretrained("musicgen-model", config=musicgen_config)( text_encoder_config: PretrainedConfig audio_encoder_config: PretrainedConfig decoder_config: MusicgenDecoderConfig **kwargs ) → MusicgenConfig
Instantiate a MusicgenConfig (or a derived class) from text encoder, audio encoder and decoder configurations.
( feature_extractor tokenizer )
Parameters
EncodecFeatureExtractor) —
An instance of EncodecFeatureExtractor. The feature extractor is a required input. T5Tokenizer) —
An instance of T5Tokenizer. The tokenizer is a required input. Constructs a MusicGen processor which wraps an EnCodec feature extractor and a T5 tokenizer into a single processor class.
MusicgenProcessor offers all the functionalities of EncodecFeatureExtractor and TTokenizer. See
__call__() and decode() for more information.
This method is used to decode either batches of audio outputs from the MusicGen model, or batches of token ids from the tokenizer. In the case of decoding token ids, this method forwards all its arguments to T5Tokenizer’s batch_decode(). Please refer to the docstring of this method for more information.
This method forwards all its arguments to T5Tokenizer’s decode(). Please refer to the docstring of this method for more information.
( config: MusicgenDecoderConfig )
Parameters
The bare Musicgen decoder model outputting raw hidden-states without any specific head on top.
The Musicgen model was proposed in Simple and Controllable Music Generation by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi, Alexandre Défossez. It is an encoder decoder transformer trained on the task of conditional music generation
This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_ids: LongTensor = None attention_mask: Optional = None encoder_hidden_states: Optional = None encoder_attention_mask: Optional = None head_mask: Optional = None cross_attn_head_mask: Optional = None past_key_values: Optional = None inputs_embeds: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None )
Parameters
torch.LongTensor of shape (batch_size * num_codebooks, sequence_length)) —
Indices of input sequence tokens in the vocabulary, corresponding to the sequence of audio codes.
Indices can be obtained by encoding an audio prompt with an audio encoder model to predict audio codes, such as with the EncodecModel. See EncodecModel.encode() for details.
The input_ids will automatically be converted from shape (batch_size * num_codebooks, target_sequence_length) to (batch_size, num_codebooks, target_sequence_length) in the forward pass. If
you obtain audio codes from an audio encoding model, such as EncodecModel, ensure that the number of
frames is equal to 1, and that you reshape the audio codes from (frames, batch_size, num_codebooks, target_sequence_length) to (batch_size * num_codebooks, target_sequence_length) prior to passing them as
input_ids.
torch.Tensor of shape (batch_size, sequence_length), optional) —
Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]:
torch.FloatTensor of shape (batch_size, encoder_sequence_length, hidden_size), optional) —
Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention of
the decoder. torch.LongTensor of shape (batch_size, encoder_sequence_length), optional) —
Mask to avoid performing cross-attention on padding tokens indices of encoder input_ids. Mask values
selected in [0, 1]:
torch.Tensor of shape (decoder_layers, decoder_attention_heads), optional) —
Mask to nullify selected heads of the attention modules. Mask values selected in [0, 1]:
torch.Tensor of shape (decoder_layers, decoder_attention_heads), optional) —
Mask to nullify selected heads of the cross-attention modules in the decoder to avoid performing
cross-attention on hidden heads. Mask values selected in [0, 1]:
tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) —
Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape
(batch_size, num_heads, sequence_length, embed_size_per_head)) and 2 additional tensors of shape
(batch_size, num_heads, encoder_sequence_length, embed_size_per_head).
Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
blocks) that can be used (see past_key_values input) to speed up sequential decoding.
If past_key_values are used, the user can optionally input only the last decoder_input_ids (those that
don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all
decoder_input_ids of shape (batch_size, sequence_length).
torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) —
Optionally, instead of passing input_ids you can choose to directly pass an embedded representation.
This is useful if you want more control over how to convert input_ids indices into associated vectors
than the model’s internal embedding lookup matrix. bool, optional) —
Whether or not to return the attentions tensors of all attention layers. See attentions under returned
tensors for more detail. bool, optional) —
Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for
more detail. bool, optional) —
Whether or not to return a ModelOutput instead of a plain tuple. The MusicgenModel forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
( config: MusicgenDecoderConfig )
Parameters
The MusicGen decoder model with a language modelling head on top.
The Musicgen model was proposed in Simple and Controllable Music Generation by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi, Alexandre Défossez. It is an encoder decoder transformer trained on the task of conditional music generation
This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_ids: LongTensor = None attention_mask: Optional = None encoder_hidden_states: Optional = None encoder_attention_mask: Optional = None head_mask: Optional = None cross_attn_head_mask: Optional = None past_key_values: Optional = None inputs_embeds: Optional = None labels: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None )
Parameters
torch.LongTensor of shape (batch_size * num_codebooks, sequence_length)) —
Indices of input sequence tokens in the vocabulary, corresponding to the sequence of audio codes.
Indices can be obtained by encoding an audio prompt with an audio encoder model to predict audio codes, such as with the EncodecModel. See EncodecModel.encode() for details.
The input_ids will automatically be converted from shape (batch_size * num_codebooks, target_sequence_length) to (batch_size, num_codebooks, target_sequence_length) in the forward pass. If
you obtain audio codes from an audio encoding model, such as EncodecModel, ensure that the number of
frames is equal to 1, and that you reshape the audio codes from (frames, batch_size, num_codebooks, target_sequence_length) to (batch_size * num_codebooks, target_sequence_length) prior to passing them as
input_ids.
torch.Tensor of shape (batch_size, sequence_length), optional) —
Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]:
torch.FloatTensor of shape (batch_size, encoder_sequence_length, hidden_size), optional) —
Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention of
the decoder. torch.LongTensor of shape (batch_size, encoder_sequence_length), optional) —
Mask to avoid performing cross-attention on padding tokens indices of encoder input_ids. Mask values
selected in [0, 1]:
torch.Tensor of shape (decoder_layers, decoder_attention_heads), optional) —
Mask to nullify selected heads of the attention modules. Mask values selected in [0, 1]:
torch.Tensor of shape (decoder_layers, decoder_attention_heads), optional) —
Mask to nullify selected heads of the cross-attention modules in the decoder to avoid performing
cross-attention on hidden heads. Mask values selected in [0, 1]:
tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) —
Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape
(batch_size, num_heads, sequence_length, embed_size_per_head)) and 2 additional tensors of shape
(batch_size, num_heads, encoder_sequence_length, embed_size_per_head).
Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
blocks) that can be used (see past_key_values input) to speed up sequential decoding.
If past_key_values are used, the user can optionally input only the last decoder_input_ids (those that
don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all
decoder_input_ids of shape (batch_size, sequence_length).
torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) —
Optionally, instead of passing input_ids you can choose to directly pass an embedded representation.
This is useful if you want more control over how to convert input_ids indices into associated vectors
than the model’s internal embedding lookup matrix. bool, optional) —
Whether or not to return the attentions tensors of all attention layers. See attentions under returned
tensors for more detail. bool, optional) —
Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for
more detail. bool, optional) —
Whether or not to return a ModelOutput instead of a plain tuple. torch.LongTensor of shape (batch_size, sequence_length), optional) —
Labels for language modeling. Note that the labels are shifted inside the model, i.e. you can set
labels = input_ids Indices are selected in [-100, 0, ..., config.vocab_size] All labels set to -100
are ignored (masked), the loss is only computed for labels in [0, ..., config.vocab_size]
Returns:
transformers.modeling_outputs.Seq2SeqLMOutput or tuple(torch.FloatTensor): A transformers.modeling_outputs.Seq2SeqLMOutput or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (MusicgenConfig) and inputs.
loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Language modeling loss.
logits (torch.FloatTensor of shape (batch_size, sequence_length, config.vocab_size)) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
past_key_values (tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) — Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape
(batch_size, num_heads, sequence_length, embed_size_per_head)) and 2 additional tensors of shape
(batch_size, num_heads, encoder_sequence_length, embed_size_per_head).
Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
blocks) that can be used (see past_key_values input) to speed up sequential decoding.
decoder_hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.FloatTensor (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).
Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
decoder_attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
cross_attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights of the decoder’s cross-attention layer, after the attention softmax, used to compute the weighted average in the cross-attention heads.
encoder_last_hidden_state (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Sequence of hidden-states at the output of the last layer of the encoder of the model.
encoder_hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.FloatTensor (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).
Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
encoder_attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
The MusicgenForCausalLM forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
( config: Optional = None text_encoder: Optional = None audio_encoder: Optional = None decoder: Optional = None )
Parameters
The composite MusicGen model with a text encoder, audio encoder and Musicgen decoder, for music generation tasks with one or both of text and audio prompts.
The Musicgen model was proposed in Simple and Controllable Music Generation by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi, Alexandre Défossez. It is an encoder decoder transformer trained on the task of conditional music generation
This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_ids: Optional = None attention_mask: Optional = None input_values: Optional = None padding_mask: Optional = None decoder_input_ids: Optional = None decoder_attention_mask: Optional = None encoder_outputs: Optional = None past_key_values: Tuple = None inputs_embeds: Optional = None decoder_inputs_embeds: Optional = None labels: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None **kwargs ) → transformers.modeling_outputs.Seq2SeqLMOutput or tuple(torch.FloatTensor)
Parameters
torch.LongTensor of shape (batch_size, sequence_length)) —
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
it.
Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
torch.Tensor of shape (batch_size, sequence_length), optional) —
Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]:
torch.LongTensor of shape (batch_size * num_codebooks, target_sequence_length), optional) —
Indices of decoder input sequence tokens in the vocabulary, corresponding to the sequence of audio codes.
Indices can be obtained by encoding an audio prompt with an audio encoder model to predict audio codes, such as with the EncodecModel. See EncodecModel.encode() for details.
The decoder_input_ids will automatically be converted from shape (batch_size * num_codebooks, target_sequence_length) to (batch_size, num_codebooks, target_sequence_length) in the forward pass. If
you obtain audio codes from an audio encoding model, such as EncodecModel, ensure that the number of
frames is equal to 1, and that you reshape the audio codes from (frames, batch_size, num_codebooks, target_sequence_length) to (batch_size * num_codebooks, target_sequence_length) prior to passing them as
decoder_input_ids.
torch.LongTensor of shape (batch_size, target_sequence_length), optional) —
Default behavior: generate a tensor that ignores pad tokens in decoder_input_ids. Causal mask will also
be used by default. torch.Tensor of shape (encoder_layers, encoder_attention_heads), optional) —
Mask to nullify selected heads of the attention modules in the encoder. Mask values selected in [0, 1]:
torch.Tensor of shape (decoder_layers, decoder_attention_heads), optional) —
Mask to nullify selected heads of the attention modules in the decoder. Mask values selected in [0, 1]:
torch.Tensor of shape (decoder_layers, decoder_attention_heads), optional) —
Mask to nullify selected heads of the cross-attention modules in the decoder. Mask values selected in [0, 1]:
tuple(tuple(torch.FloatTensor), optional) —
Tuple consists of (last_hidden_state, optional: hidden_states, optional: attentions)
last_hidden_state of shape (batch_size, sequence_length, hidden_size), optional) is a sequence of
hidden-states at the output of the last layer of the encoder. Used in the cross-attention of the decoder. tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) —
Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape
(batch_size, num_heads, sequence_length, embed_size_per_head)) and 2 additional tensors of shape
(batch_size, num_heads, encoder_sequence_length, embed_size_per_head).
Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
blocks) that can be used (see past_key_values input) to speed up sequential decoding.
If past_key_values are used, the user can optionally input only the last decoder_input_ids (those that
don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all
decoder_input_ids of shape (batch_size, sequence_length).
torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) —
Optionally, instead of passing input_ids you can choose to directly pass an embedded representation.
This is useful if you want more control over how to convert input_ids indices into associated vectors
than the model’s internal embedding lookup matrix. torch.FloatTensor of shape (batch_size, target_sequence_length, hidden_size), optional) —
Optionally, instead of passing decoder_input_ids you can choose to directly pass an embedded
representation. If past_key_values is used, optionally only the last decoder_inputs_embeds have to be
input (see past_key_values). This is useful if you want more control over how to convert
decoder_input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
If decoder_input_ids and decoder_inputs_embeds are both unset, decoder_inputs_embeds takes the value
of inputs_embeds.
bool, optional) —
If set to True, past_key_values key value states are returned and can be used to speed up decoding (see
past_key_values). bool, optional) —
Whether or not to return the attentions tensors of all attention layers. See attentions under returned
tensors for more detail. bool, optional) —
Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for
more detail. bool, optional) —
Whether or not to return a ModelOutput instead of a plain tuple. Returns
transformers.modeling_outputs.Seq2SeqLMOutput or tuple(torch.FloatTensor)
A transformers.modeling_outputs.Seq2SeqLMOutput or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (MusicgenConfig) and inputs.
loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Language modeling loss.
logits (torch.FloatTensor of shape (batch_size, sequence_length, config.vocab_size)) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
past_key_values (tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) — Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape
(batch_size, num_heads, sequence_length, embed_size_per_head)) and 2 additional tensors of shape
(batch_size, num_heads, encoder_sequence_length, embed_size_per_head).
Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
blocks) that can be used (see past_key_values input) to speed up sequential decoding.
decoder_hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.FloatTensor (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).
Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
decoder_attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
cross_attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights of the decoder’s cross-attention layer, after the attention softmax, used to compute the weighted average in the cross-attention heads.
encoder_last_hidden_state (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Sequence of hidden-states at the output of the last layer of the encoder of the model.
encoder_hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.FloatTensor (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).
Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
encoder_attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
The MusicgenForConditionalGeneration forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Examples:
>>> from transformers import AutoProcessor, MusicgenForConditionalGeneration
>>> import torch
>>> processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
>>> model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
>>> inputs = processor(
... text=["80s pop track with bassy drums and synth", "90s rock song with loud guitars and heavy drums"],
... padding=True,
... return_tensors="pt",
... )
>>> pad_token_id = model.generation_config.pad_token_id
>>> decoder_input_ids = (
... torch.ones((inputs.input_ids.shape[0] * model.decoder.num_codebooks, 1), dtype=torch.long)
... * pad_token_id
... )
>>> logits = model(**inputs, decoder_input_ids=decoder_input_ids).logits
>>> logits.shape # (bsz * num_codebooks, tgt_len, vocab_size)
torch.Size([8, 1, 2048])