Add MusicGen Melody (#28819)

* first modeling code

* make repository

* still WIP

* update model

* add tests

* add latest change

* clean docstrings and copied from

* update docstrings md and readme

* correct chroma function

* correct copied from and remove unreleated test

* add doc to toctree

* correct imports

* add convert script to notdoctested

* Add suggestion from Sanchit
Co-authored-by: Sanchit Gandhi <93869735+sanchit-gandhi@users.noreply.github.com>

* correct get_uncoditional_inputs docstrings

* modify README according to SANCHIT feedback

* add chroma to audio utils

* clean librosa and torchaudio hard dependencies

* fix FE

* refactor audio decoder -> audio encoder for consistency with previous musicgen

* refactor conditional -> encoder

* modify sampling rate logics

* modify license at the beginning

* refactor all_self_attns->all_attentions

* remove ignore copy from causallm generate

* add copied from for from_sub_models

* fix make copies

* add warning if audio is truncated

* add copied from where relevant

* remove artefact

* fix convert script

* fix torchaudio and FE

* modify chroma method according to feedback-> better naming

* refactor input_values->input_features

* refactor input_values->input_features and fix import fe

* add input_features to docstrigs

* correct inputs_embeds logics

* remove dtype conversion

* refactor _prepare_conditional_hidden_states_kwargs_for_generation ->_prepare_encoder_hidden_states_kwargs_for_generation

* change warning for chroma length

* Update src/transformers/models/musicgen_melody/convert_musicgen_melody_transformers.py
Co-authored-by: Sanchit Gandhi <93869735+sanchit-gandhi@users.noreply.github.com>

* change way to save wav, using soundfile

* correct docs and change to soundfile

* fix import

* fix init proj layers

* remove line breaks from md

* fix issue with docstrings

* add FE suggestions

* improve is in logics and remove useless imports

* remove custom from_pretrained

* simplify docstring code

* add suggestions for modeling tests

* make style

* update converting script with sanity check

* remove encoder attention mask from conditional generation

* replace musicgen melody checkpoints with official orga

* rename ylacombe->facebook in checkpoints

* fix copies

* remove unecessary warning

* add shape in code docstrings

* add files to slow doc tests

* fix md bug and add md to not_tested

* make fix-copies

* fix hidden states test and batching

---------
Co-authored-by: Sanchit Gandhi <93869735+sanchit-gandhi@users.noreply.github.com>

README.md

@@ -442,6 +442,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (from the University of Wisconsin - Madison) released with the paper [Multi Resolution Analysis (MRA) for Approximate Self-Attention](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.

README_de.md

@@ -438,6 +438,7 @@ Aktuelle Anzahl der Checkpoints: ![](https://img.shields.io/endpoint?url=https:/
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (from the University of Wisconsin - Madison) released with the paper [Multi Resolution Analysis (MRA) for Approximate Self-Attention](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.

README_es.md

@@ -415,6 +415,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (from the University of Wisconsin - Madison) released with the paper [Multi Resolution Analysis (MRA)](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.

README_fr.md

@@ -436,6 +436,7 @@ Nombre actuel de points de contrôle : ![](https://img.shields.io/endpoint?url=h
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (de l'Université du Wisconsin - Madison) a été publié dans l'article [Analyse multi-résolution (MRA) pour une auto-attention approximative](https://arxiv.org/abs/2207.10284) par Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (de Google AI) a été publié dans l'article [mT5 : un transformateur texte-à-texte pré-entraîné massivement multilingue](https://arxiv.org/abs/2010.11934) par Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (de Meta) a été publié dans l'article [Génération de musique simple et contrôlable](https://arxiv.org/abs/2306.05284) par Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi et Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (de Meta) publié dans l'article [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) parJade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (de RUC AI Box) a été publié dans l'article [MVP : Pré-entraînement supervisé multi-tâche pour la génération de langage naturel](https://arxiv.org/abs/2206.12131) par Tianyi Tang, Junyi Li, Wayne Xin Zhao et Ji-Rong Wen.
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (de SHI Labs) a été publié dans l'article [Transformateur d'attention de voisinage](https://arxiv.org/abs/2204.07143) par Ali Hassani, Steven Walton, Jiachen Li, Shen Li et Humphrey Shi.
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (du laboratoire Noah's Ark de Huawei) a été publié dans l'article [NEZHA : Représentation contextualisée neurale pour la compréhension du langage chinois](https://arxiv.org/abs/1909.00204) par Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen et Qun Liu.

README_hd.md

@@ -389,6 +389,7 @@ conda install conda-forge::transformers
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (the University of Wisconsin - Madison से) Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh. द्वाराअनुसंधान पत्र [Multi Resolution Analysis (MRA)](https://arxiv.org/abs/2207.10284) के साथ जारी किया गया
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (Google AI से) साथ वाला पेपर [mT5: एक व्यापक बहुभाषी पूर्व-प्रशिक्षित टेक्स्ट-टू-टेक्स्ट ट्रांसफॉर्मर](https://arxiv.org/abs/2010.11934) लिंटिंग ज़ू, नोआ कॉन्सटेंट, एडम रॉबर्ट्स, मिहिर काले, रामी अल-रफू, आदित्य सिद्धांत, आदित्य बरुआ, कॉलिन रैफेल द्वारा पोस्ट किया गया।
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (हुआवेई नूह के आर्क लैब से) साथ में कागज़ [NEZHA: चीनी भाषा समझ के लिए तंत्रिका प्रासंगिक प्रतिनिधित्व](https://arxiv.org/abs/1909.00204) जुन्किउ वेई, ज़ियाओज़े रेन, ज़िआओगुआंग ली, वेनयोंग हुआंग, यी लियाओ, याशेंग वांग, जियाशू लिन, शिन जियांग, जिओ चेन और कुन लियू द्वारा।

README_ja.md

@@ -449,6 +449,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (the University of Wisconsin - Madison から) Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh. から公開された研究論文 [Multi Resolution Analysis (MRA)](https://arxiv.org/abs/2207.10284)
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (Google AI から) Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel から公開された研究論文: [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934)
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (RUC AI Box から) Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen から公開された研究論文: [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131)
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (SHI Labs から) Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi から公開された研究論文: [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143)
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (Huawei Noah’s Ark Lab から) Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu から公開された研究論文: [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204)

README_ko.md

@@ -364,6 +364,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (the University of Wisconsin - Madison 에서 제공)은 Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.의 [Multi Resolution Analysis (MRA)](https://arxiv.org/abs/2207.10284) 논문과 함께 발표했습니다.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (Google AI 에서) Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel 의 [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) 논문과 함께 발표했습니다.
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (RUC AI Box 에서) Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen 의 [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) 논문과 함께 발표했습니다.
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (SHI Labs 에서) Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi 의 [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) 논문과 함께 발표했습니다.
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (Huawei Noah’s Ark Lab 에서) Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu 의 [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) 논문과 함께 발표했습니다.

README_pt-br.md

@@ -447,6 +447,7 @@ Número atual de pontos de verificação: ![](https://img.shields.io/endpoint?ur
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (from the University of Wisconsin - Madison) released with the paper [Multi Resolution Analysis (MRA) for Approximate Self-Attention](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.

README_ru.md

@@ -437,6 +437,7 @@ conda install conda-forge::transformers
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (from the University of Wisconsin - Madison) released with the paper [Multi Resolution Analysis (MRA) for Approximate Self-Attention](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.

README_te.md

@@ -439,6 +439,7 @@ Flax, PyTorch లేదా TensorFlow యొక్క ఇన్‌స్టా
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (from the University of Wisconsin - Madison) released with the paper [Multi Resolution Analysis (MRA) for Approximate Self-Attention](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.

README_vi.md

@@ -438,6 +438,7 @@ Số lượng điểm kiểm tra hiện tại: ![](https://img.shields.io/endpoi
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (từ the University of Wisconsin - Madison) được phát hành với bài báo [Multi Resolution Analysis (MRA) for Approximate Self-Attention](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (từ Google AI) được phát hành với bài báo [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (từ Meta) được phát hành với bài báo [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (từ Meta) được phát hành với bài báo [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (từ RUC AI Box) được phát hành với bài báo [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (từ SHI Labs) được phát hành với bài báo [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (từ Huawei Noah’s Ark Lab) được phát hành với bài báo [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.

README_zh-hans.md

@@ -388,6 +388,7 @@ conda install conda-forge::transformers
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (来自 the University of Wisconsin - Madison) 伴随论文 [Multi Resolution Analysis (MRA)](https://arxiv.org/abs/2207.10284) 由 Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh 发布。
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (来自 Google AI) 伴随论文 [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) 由 Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel 发布。
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (来自 中国人民大学 AI Box) 伴随论文 [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) 由 Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen 发布。
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (来自 SHI Labs) 伴随论文 [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) 由 Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi 发布。
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (来自华为诺亚方舟实验室) 伴随论文 [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) 由 Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu 发布。

README_zh-hant.md

@@ -400,6 +400,7 @@ conda install conda-forge::transformers
 1. **[MRA](https://huggingface.co/docs/transformers/model_doc/mra)** (from the University of Wisconsin - Madison) released with the paper [Multi Resolution Analysis (MRA)](https://arxiv.org/abs/2207.10284) by Zhanpeng Zeng, Sourav Pal, Jeffery Kline, Glenn M Fung, Vikas Singh.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MusicGen](https://huggingface.co/docs/transformers/model_doc/musicgen)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+1. **[MusicGen Melody](https://huggingface.co/docs/transformers/main/model_doc/musicgen_melody)** (from Meta) released with the paper [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
 1. **[NAT](https://huggingface.co/docs/transformers/model_doc/nat)** (from SHI Labs) released with the paper [Neighborhood Attention Transformer](https://arxiv.org/abs/2204.07143) by Ali Hassani, Steven Walton, Jiachen Li, Shen Li, and Humphrey Shi.
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.

docs/source/en/_toctree.yml

@@ -642,6 +642,8 @@
         title: MMS
       - local: model_doc/musicgen
         title: MusicGen
+      - local: model_doc/musicgen_melody
+        title: MusicGen Melody
       - local: model_doc/pop2piano
         title: Pop2Piano
       - local: model_doc/seamless_m4t

docs/source/en/index.md

@@ -207,6 +207,7 @@ Flax), PyTorch, and/or TensorFlow.
 |                           [MRA](model_doc/mra)                           |       ✅        |         ❌         |      ❌      |
 |                           [MT5](model_doc/mt5)                           |       ✅        |         ✅         |      ✅      |
 |                      [MusicGen](model_doc/musicgen)                      |       ✅        |         ❌         |      ❌      |
+|               [MusicGen Melody](model_doc/musicgen_melody)               |       ✅        |         ❌         |      ❌      |
 |                           [MVP](model_doc/mvp)                           |       ✅        |         ❌         |      ❌      |
 |                           [NAT](model_doc/nat)                           |       ✅        |         ❌         |      ❌      |
 |                         [Nezha](model_doc/nezha)                         |       ✅        |         ❌         |      ❌      |

docs/source/en/model_doc/musicgen_melody.md

+<!--Copyright 2024 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+
+:warning: Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# MusicGen Melody
+
+## Overview
+
+The MusicGen Melody model was proposed in [Simple and Controllable Music Generation](https://arxiv.org/abs/2306.05284) by Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi and Alexandre Défossez.
+
+MusicGen Melody 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 [ylacombe](https://huggingface.co/ylacombe). The original code can be found [here](https://github.com/facebookresearch/audiocraft). The pre-trained checkpoints can be found on the [Hugging Face Hub](https://huggingface.co/models?sort=downloads&search=facebook%2Fmusicgen).
+
+
+## Difference with [MusicGen](https://huggingface.co/docs/transformers/main/en/model_doc/musicgen)
+
+There are two key differences with MusicGen:
+1. The audio prompt is used here as a conditional signal for the generated audio sample, whereas it's used for audio continuation in [MusicGen](https://huggingface.co/docs/transformers/main/en/model_doc/musicgen).
+2. Conditional text and audio signals are concatenated to the decoder's hidden states instead of being used as a cross-attention signal, as in MusicGen.
+
+## Generation
+
+MusicGen Melody 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 [`MusicgenMelodyForConditionalGeneration.generate`], or by overriding the model's generation config (see below).
+
+Transformers supports both mono (1-channel) and stereo (2-channel) variants of MusicGen Melody. 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.
+
+
+#### Audio Conditional Generation
+
+The model can generate an audio sample conditioned on a text and an audio prompt through use of the [`MusicgenMelodyProcessor`] to pre-process the inputs.
+
+In the following examples, 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]
+```
+
+The audio file we are about to use is loaded as follows:
+```python
+>>> from datasets import load_dataset
+
+>>> dataset = load_dataset("sanchit-gandhi/gtzan", split="train", streaming=True)
+>>> sample = next(iter(dataset))["audio"]
+```
+
+The audio prompt should ideally be free of the low-frequency signals usually produced by instruments such as drums and bass. The [Demucs](https://github.com/adefossez/demucs/tree/main) model can be used to separate vocals and other signals from the drums and bass components.
+
+If you wish to use Demucs, you first need to follow the installation steps [here](https://github.com/adefossez/demucs/tree/main?tab=readme-ov-file#for-musicians) before using the following snippet:
+
+```python
+from demucs import pretrained
+from demucs.apply import apply_model
+from demucs.audio import convert_audio
+import torch
+
+
+wav = torch.tensor(sample["array"]).to(torch.float32)
+
+demucs = pretrained.get_model('htdemucs')
+
+wav = convert_audio(wav[None], sample["sampling_rate"], demucs.samplerate, demucs.audio_channels)
+wav = apply_model(demucs, wav[None])
+```
+
+You can then use the following snippet to generate music:
+
+```python
+>>> from transformers import AutoProcessor, MusicgenMelodyForConditionalGeneration
+
+>>> processor = AutoProcessor.from_pretrained("facebook/musicgen-melody")
+>>> model = MusicgenMelodyForConditionalGeneration.from_pretrained("facebook/musicgen-melody")
+
+>>> inputs = processor(
+...     audio=wav,
+...     sampling_rate=demucs.samplerate,
+...     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)
+```
+
+You can also pass the audio signal directly without using Demucs, although the quality of the generation will probably be degraded:
+
+```python
+>>> from transformers import AutoProcessor, MusicgenMelodyForConditionalGeneration
+
+>>> processor = AutoProcessor.from_pretrained("facebook/musicgen-melody")
+>>> model = MusicgenMelodyForConditionalGeneration.from_pretrained("facebook/musicgen-melody")
+
+>>> 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)
+```
+
+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:
+
+```python
+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. `soundfile`:
+
+```python
+>>> import soundfile as sf
+
+>>> sampling_rate = model.config.audio_encoder.sampling_rate
+>>> sf.write("musicgen_out.wav", audio_values[0].T.numpy(), sampling_rate)
+```
+
+
+### Text-only Conditional Generation
+
+The same [`MusicgenMelodyProcessor`] can be used to pre-process a text-only prompt. 
+
+```python
+>>> from transformers import AutoProcessor, MusicgenMelodyForConditionalGeneration
+
+>>> processor = AutoProcessor.from_pretrained("facebook/musicgen-melody")
+>>> model = MusicgenMelodyForConditionalGeneration.from_pretrained("facebook/musicgen-melody")
+
+>>> 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).
+
+
+You can also generate in batch:
+
+```python
+>>> from transformers import AutoProcessor, MusicgenMelodyForConditionalGeneration
+>>> from datasets import load_dataset
+
+>>> processor = AutoProcessor.from_pretrained("facebook/musicgen-melody")
+>>> model = MusicgenMelodyForConditionalGeneration.from_pretrained("facebook/musicgen-melody")
+
+>>> # 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)
+```
+
+### Unconditional Generation
+
+The inputs for unconditional (or 'null') generation can be obtained through the method [`MusicgenMelodyProcessor.get_unconditional_inputs`]:
+
+```python
+>>> from transformers import MusicgenMelodyForConditionalGeneration, MusicgenMelodyProcessor
+
+>>> model = MusicgenMelodyForConditionalGeneration.from_pretrained("facebook/musicgen-melody")
+>>> unconditional_inputs = MusicgenMelodyProcessor.from_pretrained("facebook/musicgen-melody").get_unconditional_inputs(num_samples=1)
+
+>>> audio_values = model.generate(**unconditional_inputs, do_sample=True, max_new_tokens=256)
+```
+
+### Generation Configuration
+
+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:
+
+```python
+>>> from transformers import MusicgenMelodyForConditionalGeneration
+
+>>> model = MusicgenMelodyForConditionalGeneration.from_pretrained("facebook/musicgen-melody")
+
+>>> # 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 = 256
+```
+
+Note 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.
+
+## Model Structure
+
+The MusicGen model can be de-composed into three distinct stages:
+1. Text encoder: maps the text inputs to a sequence of hidden-state representations. The pre-trained MusicGen models use a frozen text encoder from either T5 or Flan-T5.
+2. MusicGen Melody decoder: a language model (LM) that auto-regressively generates audio tokens (or codes) conditional on the encoder hidden-state representations
+3. Audio decoder: used to recover the audio waveform from the audio tokens predicted by the decoder.
+
+Thus, the MusicGen model can either be used as a standalone decoder model, corresponding to the class [`MusicgenMelodyForCausalLM`], or as a composite model that includes the text encoder and audio encoder, corresponding to the class [`MusicgenMelodyForConditionalGeneration`]. 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:
+
+```python
+>>> from transformers import AutoConfig, MusicgenMelodyForCausalLM, MusicgenMelodyForConditionalGeneration
+
+>>> # Option 1: get decoder config and pass to `.from_pretrained`
+>>> decoder_config = AutoConfig.from_pretrained("facebook/musicgen-melody").decoder
+>>> decoder = MusicgenMelodyForCausalLM.from_pretrained("facebook/musicgen-melody", **decoder_config.to_dict())
+
+>>> # Option 2: load the entire composite model, but only return the decoder
+>>> decoder = MusicgenMelodyForConditionalGeneration.from_pretrained("facebook/musicgen-melody").decoder
+```
+
+Since the text encoder and audio encoder models are frozen during training, the MusicGen decoder [`MusicgenMelodyForCausalLM`] 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 to recover the composite [`MusicgenMelodyForConditionalGeneration`] model.
+
+## Checkpoint Conversion
+
+- After downloading the original checkpoints from [here](https://github.com/facebookresearch/audiocraft/blob/main/docs/MUSICGEN.md#importing--exporting-models), you can convert them using the **conversion script** available at `src/transformers/models/musicgen_melody/convert_musicgen_melody_transformers.py` with the following command:
+
+```bash
+python src/transformers/models/musicgen_melody/convert_musicgen_melody_transformers.py \
+    --checkpoint="facebook/musicgen-melody" --pytorch_dump_folder /output/path 
+```
+
+Tips:
+* MusicGen is trained on the 32kHz checkpoint of Encodec. You should ensure you use a compatible version of the Encodec model.
+* Sampling mode tends to deliver better results than greedy - you can toggle sampling with the variable `do_sample` in the call to [`MusicgenMelodyForConditionalGeneration.generate`]
+
+
+## MusicgenMelodyDecoderConfig
+
+[[autodoc]] MusicgenMelodyDecoderConfig
+
+## MusicgenMelodyProcessor
+
+[[autodoc]] MusicgenMelodyProcessor
+    - get_unconditional_inputs
+
+## MusicgenMelodyFeatureExtractor
+
+[[autodoc]] MusicgenMelodyFeatureExtractor
+    - _extract_stem_indices
+
+## MusicgenMelodyConfig
+
+[[autodoc]] MusicgenMelodyConfig
+
+## MusicgenMelodyModel
+
+[[autodoc]] MusicgenMelodyModel
+    - forward
+
+## MusicgenMelodyForCausalLM
+
+[[autodoc]] MusicgenMelodyForCausalLM
+    - forward
+
+## MusicgenMelodyForConditionalGeneration
+
+[[autodoc]] MusicgenMelodyForConditionalGeneration
+    - forward
\ No newline at end of file

docs/source/en/tasks/language_modeling.md

@@ -37,7 +37,7 @@ You can finetune other architectures for causal language modeling following the
 Choose one of the following architectures:
 
 <!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->
-[BART](../model_doc/bart), [BERT](../model_doc/bert), [Bert Generation](../model_doc/bert-generation), [BigBird](../model_doc/big_bird), [BigBird-Pegasus](../model_doc/bigbird_pegasus), [BioGpt](../model_doc/biogpt), [Blenderbot](../model_doc/blenderbot), [BlenderbotSmall](../model_doc/blenderbot-small), [BLOOM](../model_doc/bloom), [CamemBERT](../model_doc/camembert), [CodeLlama](../model_doc/code_llama), [CodeGen](../model_doc/codegen), [Cohere](../model_doc/cohere), [CPM-Ant](../model_doc/cpmant), [CTRL](../model_doc/ctrl), [Data2VecText](../model_doc/data2vec-text), [ELECTRA](../model_doc/electra), [ERNIE](../model_doc/ernie), [Falcon](../model_doc/falcon), [Fuyu](../model_doc/fuyu), [Gemma](../model_doc/gemma), [GIT](../model_doc/git), [GPT-Sw3](../model_doc/gpt-sw3), [OpenAI GPT-2](../model_doc/gpt2), [GPTBigCode](../model_doc/gpt_bigcode), [GPT Neo](../model_doc/gpt_neo), [GPT NeoX](../model_doc/gpt_neox), [GPT NeoX Japanese](../model_doc/gpt_neox_japanese), [GPT-J](../model_doc/gptj), [LLaMA](../model_doc/llama), [Mamba](../model_doc/mamba), [Marian](../model_doc/marian), [mBART](../model_doc/mbart), [MEGA](../model_doc/mega), [Megatron-BERT](../model_doc/megatron-bert), [Mistral](../model_doc/mistral), [Mixtral](../model_doc/mixtral), [MPT](../model_doc/mpt), [MusicGen](../model_doc/musicgen), [MVP](../model_doc/mvp), [OpenLlama](../model_doc/open-llama), [OpenAI GPT](../model_doc/openai-gpt), [OPT](../model_doc/opt), [Pegasus](../model_doc/pegasus), [Persimmon](../model_doc/persimmon), [Phi](../model_doc/phi), [PLBart](../model_doc/plbart), [ProphetNet](../model_doc/prophetnet), [QDQBert](../model_doc/qdqbert), [Qwen2](../model_doc/qwen2), [Reformer](../model_doc/reformer), [RemBERT](../model_doc/rembert), [RoBERTa](../model_doc/roberta), [RoBERTa-PreLayerNorm](../model_doc/roberta-prelayernorm), [RoCBert](../model_doc/roc_bert), [RoFormer](../model_doc/roformer), [RWKV](../model_doc/rwkv), [Speech2Text2](../model_doc/speech_to_text_2), [StableLm](../model_doc/stablelm), [Starcoder2](../model_doc/starcoder2), [Transformer-XL](../model_doc/transfo-xl), [TrOCR](../model_doc/trocr), [Whisper](../model_doc/whisper), [XGLM](../model_doc/xglm), [XLM](../model_doc/xlm), [XLM-ProphetNet](../model_doc/xlm-prophetnet), [XLM-RoBERTa](../model_doc/xlm-roberta), [XLM-RoBERTa-XL](../model_doc/xlm-roberta-xl), [XLNet](../model_doc/xlnet), [X-MOD](../model_doc/xmod)
+[BART](../model_doc/bart), [BERT](../model_doc/bert), [Bert Generation](../model_doc/bert-generation), [BigBird](../model_doc/big_bird), [BigBird-Pegasus](../model_doc/bigbird_pegasus), [BioGpt](../model_doc/biogpt), [Blenderbot](../model_doc/blenderbot), [BlenderbotSmall](../model_doc/blenderbot-small), [BLOOM](../model_doc/bloom), [CamemBERT](../model_doc/camembert), [CodeLlama](../model_doc/code_llama), [CodeGen](../model_doc/codegen), [Cohere](../model_doc/cohere), [CPM-Ant](../model_doc/cpmant), [CTRL](../model_doc/ctrl), [Data2VecText](../model_doc/data2vec-text), [ELECTRA](../model_doc/electra), [ERNIE](../model_doc/ernie), [Falcon](../model_doc/falcon), [Fuyu](../model_doc/fuyu), [Gemma](../model_doc/gemma), [GIT](../model_doc/git), [GPT-Sw3](../model_doc/gpt-sw3), [OpenAI GPT-2](../model_doc/gpt2), [GPTBigCode](../model_doc/gpt_bigcode), [GPT Neo](../model_doc/gpt_neo), [GPT NeoX](../model_doc/gpt_neox), [GPT NeoX Japanese](../model_doc/gpt_neox_japanese), [GPT-J](../model_doc/gptj), [LLaMA](../model_doc/llama), [Mamba](../model_doc/mamba), [Marian](../model_doc/marian), [mBART](../model_doc/mbart), [MEGA](../model_doc/mega), [Megatron-BERT](../model_doc/megatron-bert), [Mistral](../model_doc/mistral), [Mixtral](../model_doc/mixtral), [MPT](../model_doc/mpt), [MusicGen](../model_doc/musicgen), [MusicGen Melody](../model_doc/musicgen_melody), [MVP](../model_doc/mvp), [OpenLlama](../model_doc/open-llama), [OpenAI GPT](../model_doc/openai-gpt), [OPT](../model_doc/opt), [Pegasus](../model_doc/pegasus), [Persimmon](../model_doc/persimmon), [Phi](../model_doc/phi), [PLBart](../model_doc/plbart), [ProphetNet](../model_doc/prophetnet), [QDQBert](../model_doc/qdqbert), [Qwen2](../model_doc/qwen2), [Reformer](../model_doc/reformer), [RemBERT](../model_doc/rembert), [RoBERTa](../model_doc/roberta), [RoBERTa-PreLayerNorm](../model_doc/roberta-prelayernorm), [RoCBert](../model_doc/roc_bert), [RoFormer](../model_doc/roformer), [RWKV](../model_doc/rwkv), [Speech2Text2](../model_doc/speech_to_text_2), [StableLm](../model_doc/stablelm), [Starcoder2](../model_doc/starcoder2), [Transformer-XL](../model_doc/transfo-xl), [TrOCR](../model_doc/trocr), [Whisper](../model_doc/whisper), [XGLM](../model_doc/xglm), [XLM](../model_doc/xlm), [XLM-ProphetNet](../model_doc/xlm-prophetnet), [XLM-RoBERTa](../model_doc/xlm-roberta), [XLM-RoBERTa-XL](../model_doc/xlm-roberta-xl), [XLNet](../model_doc/xlnet), [X-MOD](../model_doc/xmod)
 
 
 

src/transformers/__init__.py

@@ -42,6 +42,7 @@ from .utils import (
     is_timm_available,
     is_tokenizers_available,
     is_torch_available,
+    is_torchaudio_available,
     is_torchvision_available,
     is_vision_available,
     logging,
@@ -638,6 +639,11 @@ _import_structure = {
         "MusicgenConfig",
         "MusicgenDecoderConfig",
     ],
+    "models.musicgen_melody": [
+        "MUSICGEN_MELODY_PRETRAINED_MODEL_ARCHIVE_LIST",
+        "MusicgenMelodyConfig",
+        "MusicgenMelodyDecoderConfig",
+    ],
     "models.mvp": ["MvpConfig", "MvpTokenizer"],
     "models.nat": ["NAT_PRETRAINED_CONFIG_ARCHIVE_MAP", "NatConfig"],
     "models.nezha": ["NEZHA_PRETRAINED_CONFIG_ARCHIVE_MAP", "NezhaConfig"],
@@ -2787,6 +2793,15 @@ else:
             "MusicgenProcessor",
         ]
     )
+    _import_structure["models.musicgen_melody"].extend(
+        [
+            "MUSICGEN_MELODY_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "MusicgenMelodyForCausalLM",
+            "MusicgenMelodyForConditionalGeneration",
+            "MusicgenMelodyModel",
+            "MusicgenMelodyPreTrainedModel",
+        ]
+    )
     _import_structure["models.mvp"].extend(
         [
             "MVP_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -4552,6 +4567,21 @@ else:
     _import_structure["models.pop2piano"].append("Pop2PianoTokenizer")
     _import_structure["models.pop2piano"].append("Pop2PianoProcessor")
 
+try:
+    if not is_torchaudio_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from .utils import (
+        dummy_torchaudio_objects,
+    )
+
+    _import_structure["utils.dummy_torchaudio_objects"] = [
+        name for name in dir(dummy_torchaudio_objects) if not name.startswith("_")
+    ]
+else:
+    _import_structure["models.musicgen_melody"].append("MusicgenMelodyFeatureExtractor")
+    _import_structure["models.musicgen_melody"].append("MusicgenMelodyProcessor")
+
 
 # FLAX-backed objects
 try:
@@ -5459,6 +5489,11 @@ if TYPE_CHECKING:
         MusicgenConfig,
         MusicgenDecoderConfig,
     )
+    from .models.musicgen_melody import (
+        MUSICGEN_MELODY_PRETRAINED_MODEL_ARCHIVE_LIST,
+        MusicgenMelodyConfig,
+        MusicgenMelodyDecoderConfig,
+    )
     from .models.mvp import MvpConfig, MvpTokenizer
     from .models.nat import NAT_PRETRAINED_CONFIG_ARCHIVE_MAP, NatConfig
     from .models.nezha import NEZHA_PRETRAINED_CONFIG_ARCHIVE_MAP, NezhaConfig
@@ -7360,6 +7395,13 @@ if TYPE_CHECKING:
             MusicgenPreTrainedModel,
             MusicgenProcessor,
         )
+        from .models.musicgen_melody import (
+            MUSICGEN_MELODY_PRETRAINED_MODEL_ARCHIVE_LIST,
+            MusicgenMelodyForCausalLM,
+            MusicgenMelodyForConditionalGeneration,
+            MusicgenMelodyModel,
+            MusicgenMelodyPreTrainedModel,
+        )
         from .models.mvp import (
             MVP_PRETRAINED_MODEL_ARCHIVE_LIST,
             MvpForCausalLM,
@@ -8811,6 +8853,13 @@ if TYPE_CHECKING:
             Pop2PianoTokenizer,
         )
 
+    try:
+        if not is_torchaudio_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from .utils.dummy_torchaudio_objects import *
+    else:
+        from .models.musicgen_melody import MusicgenMelodyFeatureExtractor, MusicgenMelodyProcessor
     try:
         if not is_flax_available():
             raise OptionalDependencyNotAvailable()

src/transformers/audio_utils.py

@@ -17,7 +17,7 @@ Audio processing functions to extract features from audio waveforms. This code i
 and remove unnecessary dependencies.
 """
 import warnings
-from typing import Optional, Union
+from typing import Optional, Tuple, Union
 
 import numpy as np
 
@@ -94,6 +94,29 @@ def mel_to_hertz(mels: Union[float, np.ndarray], mel_scale: str = "htk") -> Unio
     return freq
 
 
+def hertz_to_octave(
+    freq: Union[float, np.ndarray], tuning: Optional[float] = 0.0, bins_per_octave: Optional[int] = 12
+):
+    """
+    Convert frequency from hertz to fractional octave numbers.
+    Adapted from *librosa*.
+
+    Args:
+        freq (`float` or `np.ndarray`):
+            The frequency, or multiple frequencies, in hertz (Hz).
+        tuning (`float`, defaults to `0.`):
+            Tuning deviation from the Stuttgart pitch (A440) in (fractional) bins per octave.
+        bins_per_octave (`int`, defaults to `12`):
+            Number of bins per octave.
+
+    Returns:
+        `float` or `np.ndarray`: The frequencies on the octave scale.
+    """
+    stuttgart_pitch = 440.0 * 2.0 ** (tuning / bins_per_octave)
+    octave = np.log2(freq / (float(stuttgart_pitch) / 16))
+    return octave
+
+
 def _create_triangular_filter_bank(fft_freqs: np.ndarray, filter_freqs: np.ndarray) -> np.ndarray:
     """
     Creates a triangular filter bank.
@@ -116,6 +139,81 @@ def _create_triangular_filter_bank(fft_freqs: np.ndarray, filter_freqs: np.ndarr
     return np.maximum(np.zeros(1), np.minimum(down_slopes, up_slopes))
 
 
+def chroma_filter_bank(
+    num_frequency_bins: int,
+    num_chroma: int,
+    sampling_rate: int,
+    tuning: float = 0.0,
+    power: Optional[float] = 2.0,
+    weighting_parameters: Optional[Tuple[float]] = (5.0, 2),
+    start_at_c_chroma: Optional[bool] = True,
+):
+    """
+    Creates a chroma filter bank, i.e a linear transformation to project spectrogram bins onto chroma bins.
+
+    Adapted from *librosa*.
+
+    Args:
+        num_frequency_bins (`int`):
+            Number of frequencies used to compute the spectrogram (should be the same as in `stft`).
+        num_chroma (`int`):
+            Number of chroma bins (i.e pitch classes).
+        sampling_rate (`float`):
+            Sample rate of the audio waveform.
+        tuning (`float`):
+            Tuning deviation from A440 in fractions of a chroma bin.
+        power (`float`, *optional*, defaults to 2.0):
+            If 12.0, normalizes each column with their L2 norm. If 1.0, normalizes each column with their L1 norm.
+        weighting_parameters (`Tuple[float]`, *optional*, defaults to `(5., 2.)`):
+            If specified, apply a Gaussian weighting parameterized by the first element of the tuple being the center and
+            the second element being the Gaussian half-width.
+        start_at_c_chroma (`float`, *optional*, defaults to `True`):
+            If True, the filter bank will start at the 'C' pitch class. Otherwise, it will start at 'A'.
+    Returns:
+        `np.ndarray` of shape `(num_frequency_bins, num_chroma)`
+    """
+    # Get the FFT bins, not counting the DC component
+    frequencies = np.linspace(0, sampling_rate, num_frequency_bins, endpoint=False)[1:]
+
+    freq_bins = num_chroma * hertz_to_octave(frequencies, tuning=tuning, bins_per_octave=num_chroma)
+
+    # make up a value for the 0 Hz bin = 1.5 octaves below bin 1
+    # (so chroma is 50% rotated from bin 1, and bin width is broad)
+    freq_bins = np.concatenate(([freq_bins[0] - 1.5 * num_chroma], freq_bins))
+
+    bins_width = np.concatenate((np.maximum(freq_bins[1:] - freq_bins[:-1], 1.0), [1]))
+
+    chroma_filters = np.subtract.outer(freq_bins, np.arange(0, num_chroma, dtype="d")).T
+
+    num_chroma2 = np.round(float(num_chroma) / 2)
+
+    # Project into range -num_chroma/2 .. num_chroma/2
+    # add on fixed offset of 10*num_chroma to ensure all values passed to
+    # rem are positive
+    chroma_filters = np.remainder(chroma_filters + num_chroma2 + 10 * num_chroma, num_chroma) - num_chroma2
+
+    # Gaussian bumps - 2*D to make them narrower
+    chroma_filters = np.exp(-0.5 * (2 * chroma_filters / np.tile(bins_width, (num_chroma, 1))) ** 2)
+
+    # normalize each column
+    if power is not None:
+        chroma_filters = chroma_filters / np.sum(chroma_filters**power, axis=0, keepdims=True) ** (1.0 / power)
+
+    # Maybe apply scaling for fft bins
+    if weighting_parameters is not None:
+        center, half_width = weighting_parameters
+        chroma_filters *= np.tile(
+            np.exp(-0.5 * (((freq_bins / num_chroma - center) / half_width) ** 2)),
+            (num_chroma, 1),
+        )
+
+    if start_at_c_chroma:
+        chroma_filters = np.roll(chroma_filters, -3 * (num_chroma // 12), axis=0)
+
+    # remove aliasing columns, copy to ensure row-contiguity
+    return np.ascontiguousarray(chroma_filters[:, : int(1 + num_frequency_bins / 2)])
+
+
 def mel_filter_bank(
     num_frequency_bins: int,
     num_mel_filters: int,

src/transformers/models/__init__.py

@@ -153,6 +153,7 @@ from . import (
     mra,
     mt5,
     musicgen,
+    musicgen_melody,
     mvp,
     nat,
     nezha,