Add Flash Attention 2 support to Musicgen and Musicgen Melody (#29939)

* add FA2 to o.g Musicgen

* make style

* add FA2 support to Musicgen Melody

* add generation FA2 tests to o.g Musicgen

* make style and fix copies

* add Musicgen to FA2 docs + deprecate list

* add sdpa supports to Musicgen's

* make style and fix copies

* refactor attention implementation arguments

* add Copied from to sdpa tests

* add copied form in sdpa tests melody

* add copied for FA2 generation tests

* add FA2 inference copied from

* make style

docs/source/en/perf_infer_gpu_one.md

@@ -55,6 +55,8 @@ FlashAttention-2 is currently supported for the following architectures:
 * [MBart](https://huggingface.co/docs/transformers/model_doc/mbart#transformers.MBartModel)
 * [Mistral](https://huggingface.co/docs/transformers/model_doc/mistral#transformers.MistralModel)
 * [Mixtral](https://huggingface.co/docs/transformers/model_doc/mixtral#transformers.MixtralModel)
+* [Musicgen](https://huggingface.co/docs/transformers/model_doc/musicgen#transformers.MusicgenModel)
+* [MusicGen Melody](https://huggingface.co/docs/transformers/model_doc/musicgen_melody#transformers.MusicgenMelodyModel)
 * [OPT](https://huggingface.co/docs/transformers/model_doc/opt#transformers.OPTModel)
 * [Phi](https://huggingface.co/docs/transformers/model_doc/phi#transformers.PhiModel)
 * [StableLm](https://huggingface.co/docs/transformers/model_doc/stablelm#transformers.StableLmModel)
@@ -190,6 +192,8 @@ For now, Transformers supports SDPA inference and training for the following arc
 * [Starcoder2](https://huggingface.co/docs/transformers/model_doc/starcoder2#transformers.Starcoder2Model)
 * [Qwen2](https://huggingface.co/docs/transformers/model_doc/qwen2#transformers.Qwen2Model)
 * [Qwen2MoE](https://huggingface.co/docs/transformers/model_doc/qwen2_moe#transformers.Qwen2MoeModel)
+* [Musicgen](https://huggingface.co/docs/transformers/model_doc/musicgen#transformers.MusicgenModel)
+* [MusicGen Melody](https://huggingface.co/docs/transformers/model_doc/musicgen_melody#transformers.MusicgenMelodyModel)
 
 <Tip>
 

src/transformers/models/deprecated/_archive_maps.py

@@ -1470,6 +1470,12 @@ MUSICGEN_PRETRAINED_CONFIG_ARCHIVE_MAP = DeprecatedDict(
 
 MUSICGEN_PRETRAINED_MODEL_ARCHIVE_LIST = DeprecatedList(["facebook/musicgen-small"])
 
+MUSICGEN_MELODY_PRETRAINED_CONFIG_ARCHIVE_MAP = DeprecatedDict(
+    {"facebook/musicgen-melody": "https://huggingface.co/facebook/musicgen-melody/resolve/main/config.json"}
+)
+
+MUSICGEN_MELODY_PRETRAINED_MODEL_ARCHIVE_LIST = DeprecatedList(["facebook/musicgen-melody"])
+
 MVP_PRETRAINED_MODEL_ARCHIVE_LIST = DeprecatedList(
     [
         "RUCAIBox/mvp",

src/transformers/models/musicgen/configuration_musicgen.py

@@ -239,3 +239,20 @@ class MusicgenConfig(PretrainedConfig):
     # This is a property because you might want to change the codec model on the fly
     def sampling_rate(self):
         return self.audio_encoder.sampling_rate
+
+    @property
+    def _attn_implementation(self):
+        # This property is made private for now (as it cannot be changed and a PreTrainedModel.use_attn_implementation method needs to be implemented.)
+        if hasattr(self, "_attn_implementation_internal"):
+            if self._attn_implementation_internal is None:
+                # `config.attn_implementation` should never be None, for backward compatibility.
+                return "eager"
+            else:
+                return self._attn_implementation_internal
+        else:
+            return "eager"
+
+    @_attn_implementation.setter
+    def _attn_implementation(self, value):
+        self._attn_implementation_internal = value
+        self.decoder._attn_implementation = value

src/transformers/models/musicgen/modeling_musicgen.py

@@ -22,13 +22,19 @@ from typing import TYPE_CHECKING, Any, Dict, Optional, Tuple, Union
 
 import torch
 import torch.nn as nn
+import torch.nn.functional as F
 from torch.nn import CrossEntropyLoss
 
 from ...activations import ACT2FN
 from ...generation.configuration_utils import GenerationConfig
 from ...generation.logits_process import ClassifierFreeGuidanceLogitsProcessor, LogitsProcessorList
 from ...generation.stopping_criteria import StoppingCriteriaList
-from ...modeling_attn_mask_utils import _prepare_4d_attention_mask, _prepare_4d_causal_attention_mask
+from ...modeling_attn_mask_utils import (
+    _prepare_4d_attention_mask,
+    _prepare_4d_attention_mask_for_sdpa,
+    _prepare_4d_causal_attention_mask,
+    _prepare_4d_causal_attention_mask_for_sdpa,
+)
 from ...modeling_outputs import (
     BaseModelOutput,
     BaseModelOutputWithPastAndCrossAttentions,
@@ -40,6 +46,8 @@ from ...modeling_utils import PreTrainedModel
 from ...utils import (
     add_start_docstrings,
     add_start_docstrings_to_model_forward,
+    is_flash_attn_2_available,
+    is_flash_attn_greater_or_equal_2_10,
     logging,
     replace_return_docstrings,
 )
@@ -48,6 +56,10 @@ from ..auto.modeling_auto import AutoModel
 from .configuration_musicgen import MusicgenConfig, MusicgenDecoderConfig
 
 
+if is_flash_attn_2_available():
+    from flash_attn import flash_attn_func, flash_attn_varlen_func
+    from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input  # noqa
+
 if TYPE_CHECKING:
     from ...generation.streamers import BaseStreamer
 
@@ -60,6 +72,19 @@ _CHECKPOINT_FOR_DOC = "facebook/musicgen-small"
 from ..deprecated._archive_maps import MUSICGEN_PRETRAINED_MODEL_ARCHIVE_LIST  # noqa: F401, E402
 
 
+# Copied from transformers.models.llama.modeling_llama._get_unpad_data
+def _get_unpad_data(attention_mask):
+    seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
+    indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
+    max_seqlen_in_batch = seqlens_in_batch.max().item()
+    cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
+    return (
+        indices,
+        cu_seqlens,
+        max_seqlen_in_batch,
+    )
+
+
 @dataclass
 class MusicgenUnconditionalInput(ModelOutput):
     """
@@ -302,29 +327,361 @@ class MusicgenAttention(nn.Module):
         return attn_output, attn_weights_reshaped, past_key_value
 
 
+# Copied from transformers.models.bart.modeling_bart.BartFlashAttention2 with Bart->Musicgen
+class MusicgenFlashAttention2(MusicgenAttention):
+    """
+    Musicgen flash attention module. This module inherits from `MusicgenAttention` as the weights of the module stays
+    untouched. The only required change would be on the forward pass where it needs to correctly call the public API of
+    flash attention and deal with padding tokens in case the input contains any of them.
+    """
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2.__init__
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
+        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
+        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
+        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
+
+    def _reshape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        key_value_states: Optional[torch.Tensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        layer_head_mask: Optional[torch.Tensor] = None,
+        output_attentions: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        # MusicgenFlashAttention2 attention does not support output_attentions
+        if output_attentions:
+            raise ValueError("MusicgenFlashAttention2 attention does not support output_attentions")
+
+        # if key_value_states are provided this layer is used as a cross-attention layer
+        # for the decoder
+        is_cross_attention = key_value_states is not None
+
+        bsz, q_len, _ = hidden_states.size()
+
+        # get query proj
+        query_states = self._reshape(self.q_proj(hidden_states), -1, bsz)
+        # get key, value proj
+        # `past_key_value[0].shape[2] == key_value_states.shape[1]`
+        # is checking that the `sequence_length` of the `past_key_value` is the same as
+        # the provided `key_value_states` to support prefix tuning
+        if (
+            is_cross_attention
+            and past_key_value is not None
+            and past_key_value[0].shape[2] == key_value_states.shape[1]
+        ):
+            # reuse k,v, cross_attentions
+            key_states = past_key_value[0].transpose(1, 2)
+            value_states = past_key_value[1].transpose(1, 2)
+        elif is_cross_attention:
+            # cross_attentions
+            key_states = self._reshape(self.k_proj(key_value_states), -1, bsz)
+            value_states = self._reshape(self.v_proj(key_value_states), -1, bsz)
+        elif past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = self._reshape(self.k_proj(hidden_states), -1, bsz)
+            value_states = self._reshape(self.v_proj(hidden_states), -1, bsz)
+            key_states = torch.cat([past_key_value[0].transpose(1, 2), key_states], dim=1)
+            value_states = torch.cat([past_key_value[1].transpose(1, 2), value_states], dim=1)
+        else:
+            # self_attention
+            key_states = self._reshape(self.k_proj(hidden_states), -1, bsz)
+            value_states = self._reshape(self.v_proj(hidden_states), -1, bsz)
+
+        if self.is_decoder:
+            # if cross_attention save Tuple(torch.Tensor, torch.Tensor) of all cross attention key/value_states.
+            # Further calls to cross_attention layer can then reuse all cross-attention
+            # key/value_states (first "if" case)
+            # if uni-directional self-attention (decoder) save Tuple(torch.Tensor, torch.Tensor) of
+            # all previous decoder key/value_states. Further calls to uni-directional self-attention
+            # can concat previous decoder key/value_states to current projected key/value_states (third "elif" case)
+            # if encoder bi-directional self-attention `past_key_value` is always `None`
+            past_key_value = (key_states.transpose(1, 2), value_states.transpose(1, 2))
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value[0].shape[-2]
+
+        # In PEFT, usually we cast the layer norms in float32 for training stability reasons
+        # therefore the input hidden states gets silently casted in float32. Hence, we need
+        # cast them back in the correct dtype just to be sure everything works as expected.
+        # This might slowdown training & inference so it is recommended to not cast the LayerNorms
+        # in fp32. (LlamaRMSNorm handles it correctly)
+
+        input_dtype = query_states.dtype
+        if input_dtype == torch.float32:
+            if torch.is_autocast_enabled():
+                target_dtype = torch.get_autocast_gpu_dtype()
+            # Handle the case where the model is quantized
+            elif hasattr(self.config, "_pre_quantization_dtype"):
+                target_dtype = self.config._pre_quantization_dtype
+            else:
+                target_dtype = self.q_proj.weight.dtype
+
+            logger.warning_once(
+                f"The input hidden states seems to be silently casted in float32, this might be related to"
+                f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
+                f" {target_dtype}."
+            )
+
+            query_states = query_states.to(target_dtype)
+            key_states = key_states.to(target_dtype)
+            value_states = value_states.to(target_dtype)
+
+        attn_output = self._flash_attention_forward(
+            query_states, key_states, value_states, attention_mask, q_len, dropout=self.dropout
+        )
+
+        attn_output = attn_output.reshape(bsz, q_len, -1)
+        attn_output = self.out_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._flash_attention_forward
+    def _flash_attention_forward(
+        self, query_states, key_states, value_states, attention_mask, query_length, dropout=0.0, softmax_scale=None
+    ):
+        """
+        Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
+        first unpad the input, then computes the attention scores and pad the final attention scores.
+        Args:
+            query_states (`torch.Tensor`):
+                Input query states to be passed to Flash Attention API
+            key_states (`torch.Tensor`):
+                Input key states to be passed to Flash Attention API
+            value_states (`torch.Tensor`):
+                Input value states to be passed to Flash Attention API
+            attention_mask (`torch.Tensor`):
+                The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
+                position of padding tokens and 1 for the position of non-padding tokens.
+            dropout (`float`):
+                Attention dropout
+            softmax_scale (`float`, *optional*):
+                The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
+        """
+        if not self._flash_attn_uses_top_left_mask:
+            causal = self.is_causal
+        else:
+            # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
+            causal = self.is_causal and query_length != 1
+
+        # Contains at least one padding token in the sequence
+        if attention_mask is not None:
+            batch_size = query_states.shape[0]
+            query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
+                query_states, key_states, value_states, attention_mask, query_length
+            )
+
+            cu_seqlens_q, cu_seqlens_k = cu_seq_lens
+            max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
+
+            attn_output_unpad = flash_attn_varlen_func(
+                query_states,
+                key_states,
+                value_states,
+                cu_seqlens_q=cu_seqlens_q,
+                cu_seqlens_k=cu_seqlens_k,
+                max_seqlen_q=max_seqlen_in_batch_q,
+                max_seqlen_k=max_seqlen_in_batch_k,
+                dropout_p=dropout,
+                softmax_scale=softmax_scale,
+                causal=causal,
+            )
+
+            attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
+        else:
+            attn_output = flash_attn_func(
+                query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal
+            )
+
+        return attn_output
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._upad_input
+    def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
+        indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
+        batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
+
+        key_layer = index_first_axis(
+            key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
+        )
+        value_layer = index_first_axis(
+            value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
+        )
+        if query_length == kv_seq_len:
+            query_layer = index_first_axis(
+                query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim), indices_k
+            )
+            cu_seqlens_q = cu_seqlens_k
+            max_seqlen_in_batch_q = max_seqlen_in_batch_k
+            indices_q = indices_k
+        elif query_length == 1:
+            max_seqlen_in_batch_q = 1
+            cu_seqlens_q = torch.arange(
+                batch_size + 1, dtype=torch.int32, device=query_layer.device
+            )  # There is a memcpy here, that is very bad.
+            indices_q = cu_seqlens_q[:-1]
+            query_layer = query_layer.squeeze(1)
+        else:
+            # The -q_len: slice assumes left padding.
+            attention_mask = attention_mask[:, -query_length:]
+            query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
+
+        return (
+            query_layer,
+            key_layer,
+            value_layer,
+            indices_q,
+            (cu_seqlens_q, cu_seqlens_k),
+            (max_seqlen_in_batch_q, max_seqlen_in_batch_k),
+        )
+
+
+# Copied from transformers.models.bart.modeling_bart.BartSdpaAttention with Bart->Musicgen
+class MusicgenSdpaAttention(MusicgenAttention):
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        key_value_states: Optional[torch.Tensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        layer_head_mask: Optional[torch.Tensor] = None,
+        output_attentions: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        """Input shape: Batch x Time x Channel"""
+        if output_attentions or layer_head_mask is not None:
+            # TODO: Improve this warning with e.g. `model.config._attn_implementation = "manual"` once this is implemented.
+            logger.warning_once(
+                "MusicgenModel is using MusicgenSdpaAttention, but `torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True` or `layer_head_mask` not None. Falling back to the manual attention"
+                ' implementation, but specifying the manual implementation will be required from Transformers version v5.0.0 onwards. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
+            )
+            return super().forward(
+                hidden_states,
+                key_value_states=key_value_states,
+                past_key_value=past_key_value,
+                attention_mask=attention_mask,
+                layer_head_mask=layer_head_mask,
+                output_attentions=output_attentions,
+            )
+
+        # if key_value_states are provided this layer is used as a cross-attention layer
+        # for the decoder
+        is_cross_attention = key_value_states is not None
+
+        bsz, tgt_len, _ = hidden_states.size()
+
+        # get query proj
+        query_states = self.q_proj(hidden_states)
+        # get key, value proj
+        # `past_key_value[0].shape[2] == key_value_states.shape[1]`
+        # is checking that the `sequence_length` of the `past_key_value` is the same as
+        # the provided `key_value_states` to support prefix tuning
+        if (
+            is_cross_attention
+            and past_key_value is not None
+            and past_key_value[0].shape[2] == key_value_states.shape[1]
+        ):
+            # reuse k,v, cross_attentions
+            key_states = past_key_value[0]
+            value_states = past_key_value[1]
+        elif is_cross_attention:
+            # cross_attentions
+            key_states = self._shape(self.k_proj(key_value_states), -1, bsz)
+            value_states = self._shape(self.v_proj(key_value_states), -1, bsz)
+        elif past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
+            value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
+            key_states = torch.cat([past_key_value[0], key_states], dim=2)
+            value_states = torch.cat([past_key_value[1], value_states], dim=2)
+        else:
+            # self_attention
+            key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
+            value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
+
+        if self.is_decoder:
+            # if cross_attention save Tuple(torch.Tensor, torch.Tensor) of all cross attention key/value_states.
+            # Further calls to cross_attention layer can then reuse all cross-attention
+            # key/value_states (first "if" case)
+            # if uni-directional self-attention (decoder) save Tuple(torch.Tensor, torch.Tensor) of
+            # all previous decoder key/value_states. Further calls to uni-directional self-attention
+            # can concat previous decoder key/value_states to current projected key/value_states (third "elif" case)
+            # if encoder bi-directional self-attention `past_key_value` is always `None`
+            past_key_value = (key_states, value_states)
+
+        query_states = self._shape(query_states, tgt_len, bsz)
+
+        # NOTE: SDPA with memory-efficient backend is currently (torch==2.1.2) bugged when using non-contiguous inputs and a custom attn_mask,
+        # but we are fine here as `_shape` do call `.contiguous()`. Reference: https://github.com/pytorch/pytorch/issues/112577
+        attn_output = torch.nn.functional.scaled_dot_product_attention(
+            query_states,
+            key_states,
+            value_states,
+            attn_mask=attention_mask,
+            dropout_p=self.dropout if self.training else 0.0,
+            # The tgt_len > 1 is necessary to match with AttentionMaskConverter.to_causal_4d that does not create a causal mask in case tgt_len == 1.
+            is_causal=self.is_causal and attention_mask is None and tgt_len > 1,
+        )
+
+        if attn_output.size() != (bsz, self.num_heads, tgt_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, tgt_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.transpose(1, 2)
+
+        # Use the `embed_dim` from the config (stored in the class) rather than `hidden_state` because `attn_output` can be
+        # partitioned across GPUs when using tensor-parallelism.
+        attn_output = attn_output.reshape(bsz, tgt_len, self.embed_dim)
+
+        attn_output = self.out_proj(attn_output)
+
+        return attn_output, None, past_key_value
+
+
+MUSICGEN_ATTENTION_CLASSES = {
+    "eager": MusicgenAttention,
+    "sdpa": MusicgenSdpaAttention,
+    "flash_attention_2": MusicgenFlashAttention2,
+}
+
+
 class MusicgenDecoderLayer(nn.Module):
     def __init__(self, config: MusicgenDecoderConfig):
         super().__init__()
         self.embed_dim = config.hidden_size
 
-        self.self_attn = MusicgenAttention(
+        self.self_attn = MUSICGEN_ATTENTION_CLASSES[config._attn_implementation](
             embed_dim=self.embed_dim,
             num_heads=config.num_attention_heads,
             dropout=config.attention_dropout,
             is_decoder=True,
             bias=False,
+            is_causal=True,
+            config=config,
         )
         self.dropout = config.dropout
         self.activation_fn = ACT2FN[config.activation_function]
         self.activation_dropout = config.activation_dropout
 
         self.self_attn_layer_norm = nn.LayerNorm(self.embed_dim)
-        self.encoder_attn = MusicgenAttention(
+        self.encoder_attn = MUSICGEN_ATTENTION_CLASSES[config._attn_implementation](
             self.embed_dim,
             config.num_attention_heads,
             dropout=config.attention_dropout,
             is_decoder=True,
             bias=False,
+            config=config,
         )
         self.encoder_attn_layer_norm = nn.LayerNorm(self.embed_dim)
         self.fc1 = nn.Linear(self.embed_dim, config.ffn_dim, bias=False)
@@ -432,6 +789,8 @@ class MusicgenPreTrainedModel(PreTrainedModel):
     base_model_prefix = "model"
     supports_gradient_checkpointing = True
     _no_split_modules = ["MusicgenDecoderLayer", "MusicgenAttention"]
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
 
     def _init_weights(self, module):
         std = self.config.initializer_factor
@@ -667,6 +1026,7 @@ class MusicgenDecoder(MusicgenPreTrainedModel):
 
         self.layers = nn.ModuleList([MusicgenDecoderLayer(config) for _ in range(config.num_hidden_layers)])
         self.layer_norm = nn.LayerNorm(config.hidden_size)
+        self.attn_implementation = config._attn_implementation
 
         self.gradient_checkpointing = False
         # Initialize weights and apply final processing
@@ -721,16 +1081,40 @@ class MusicgenDecoder(MusicgenPreTrainedModel):
         if inputs_embeds is None:
             inputs_embeds = sum([self.embed_tokens[codebook](input[:, codebook]) for codebook in range(num_codebooks)])
 
-        attention_mask = _prepare_4d_causal_attention_mask(
+        if self.attn_implementation == "flash_attention_2":
-            attention_mask, input_shape, inputs_embeds, past_key_values_length
+            attention_mask = attention_mask if (attention_mask is not None and 0 in attention_mask) else None
-        )
+        elif self.attn_implementation == "sdpa" and head_mask is None and not output_attentions:
+            # output_attentions=True & cross_attn_head_mask can not be supported when using SDPA, and we fall back on
+            # the manual implementation that requires a 4D causal mask in all cases.
+            attention_mask = _prepare_4d_causal_attention_mask_for_sdpa(
+                attention_mask,
+                input_shape,
+                inputs_embeds,
+                past_key_values_length,
+            )
+        else:
+            attention_mask = _prepare_4d_causal_attention_mask(
+                attention_mask, input_shape, inputs_embeds, past_key_values_length
+            )
 
         # expand encoder attention mask
         if encoder_hidden_states is not None and encoder_attention_mask is not None:
-            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+            if self.attn_implementation == "flash_attention_2":
-            encoder_attention_mask = _prepare_4d_attention_mask(
+                encoder_attention_mask = encoder_attention_mask if 0 in encoder_attention_mask else None
-                encoder_attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]
+            elif self.attn_implementation == "sdpa" and cross_attn_head_mask is None and not output_attentions:
-            )
+                # output_attentions=True & cross_attn_head_mask can not be supported when using SDPA, and we fall back on
+                # the manual implementation that requires a 4D causal mask in all cases.
+                # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+                encoder_attention_mask = _prepare_4d_attention_mask_for_sdpa(
+                    encoder_attention_mask,
+                    inputs_embeds.dtype,
+                    tgt_len=input_shape[-1],
+                )
+            else:
+                # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+                encoder_attention_mask = _prepare_4d_attention_mask(
+                    encoder_attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]
+                )
 
         # embed positions
         positions = self.embed_positions(input, past_key_values_length)
@@ -1409,6 +1793,8 @@ class MusicgenForConditionalGeneration(PreTrainedModel):
     base_model_prefix = "encoder_decoder"
     main_input_name = "input_ids"
     supports_gradient_checkpointing = True
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
 
     def __init__(
         self,

src/transformers/models/musicgen_melody/configuration_musicgen_melody.py

@@ -21,9 +21,7 @@ from ..auto.configuration_auto import AutoConfig
 
 logger = logging.get_logger(__name__)
 
-MUSICGEN_MELODY_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+from ..deprecated._archive_maps import MUSICGEN_MELODY_PRETRAINED_CONFIG_ARCHIVE_MAP  # noqa: F401, E402
-    "facebook/musicgen-melody": "https://huggingface.co/facebook/musicgen-melody/resolve/main/config.json",
-}
 
 
 class MusicgenMelodyDecoderConfig(PretrainedConfig):
@@ -254,3 +252,20 @@ class MusicgenMelodyConfig(PretrainedConfig):
     # This is a property because you might want to change the codec model on the fly
     def sampling_rate(self):
         return self.audio_encoder.sampling_rate
+
+    @property
+    def _attn_implementation(self):
+        # This property is made private for now (as it cannot be changed and a PreTrainedModel.use_attn_implementation method needs to be implemented.)
+        if hasattr(self, "_attn_implementation_internal"):
+            if self._attn_implementation_internal is None:
+                # `config.attn_implementation` should never be None, for backward compatibility.
+                return "eager"
+            else:
+                return self._attn_implementation_internal
+        else:
+            return "eager"
+
+    @_attn_implementation.setter
+    def _attn_implementation(self, value):
+        self._attn_implementation_internal = value
+        self.decoder._attn_implementation = value

src/transformers/models/musicgen_melody/modeling_musicgen_melody.py

@@ -22,13 +22,14 @@ from typing import TYPE_CHECKING, Any, Dict, Optional, Tuple, Union
 
 import torch
 import torch.nn as nn
+import torch.nn.functional as F
 from torch.nn import CrossEntropyLoss
 
 from ...activations import ACT2FN
 from ...generation.configuration_utils import GenerationConfig
 from ...generation.logits_process import ClassifierFreeGuidanceLogitsProcessor, LogitsProcessorList
 from ...generation.stopping_criteria import StoppingCriteriaList
-from ...modeling_attn_mask_utils import _prepare_4d_causal_attention_mask
+from ...modeling_attn_mask_utils import _prepare_4d_causal_attention_mask, _prepare_4d_causal_attention_mask_for_sdpa
 from ...modeling_outputs import (
     BaseModelOutputWithPast,
     ModelOutput,
@@ -37,6 +38,8 @@ from ...modeling_utils import PreTrainedModel
 from ...utils import (
     add_start_docstrings,
     add_start_docstrings_to_model_forward,
+    is_flash_attn_2_available,
+    is_flash_attn_greater_or_equal_2_10,
     logging,
     replace_return_docstrings,
 )
@@ -45,6 +48,10 @@ from ..auto.modeling_auto import AutoModel, AutoModelForTextEncoding
 from .configuration_musicgen_melody import MusicgenMelodyConfig, MusicgenMelodyDecoderConfig
 
 
+if is_flash_attn_2_available():
+    from flash_attn import flash_attn_func, flash_attn_varlen_func
+    from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input  # noqa
+
 if TYPE_CHECKING:
     from ...generation.streamers import BaseStreamer
 
@@ -53,10 +60,20 @@ logger = logging.get_logger(__name__)
 _CONFIG_FOR_DOC = "MusicgenMelodyConfig"
 _CHECKPOINT_FOR_DOC = "facebook/musicgen-melody"
 
-MUSICGEN_MELODY_PRETRAINED_MODEL_ARCHIVE_LIST = [
+from ..deprecated._archive_maps import MUSICGEN_MELODY_PRETRAINED_MODEL_ARCHIVE_LIST  # noqa: F401, E402
-    "facebook/musicgen-melody",
+
-    # See all Musicgen Melody models at https://huggingface.co/models?filter=musicgen_melody
+
-]
+# Copied from transformers.models.llama.modeling_llama._get_unpad_data
+def _get_unpad_data(attention_mask):
+    seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
+    indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
+    max_seqlen_in_batch = seqlens_in_batch.max().item()
+    cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
+    return (
+        indices,
+        cu_seqlens,
+        max_seqlen_in_batch,
+    )
 
 
 @dataclass
@@ -324,17 +341,348 @@ class MusicgenMelodyAttention(nn.Module):
         return attn_output, attn_weights_reshaped, past_key_value
 
 
+# Copied from transformers.models.bart.modeling_bart.BartFlashAttention2 with Bart->MusicgenMelody
+class MusicgenMelodyFlashAttention2(MusicgenMelodyAttention):
+    """
+    MusicgenMelody flash attention module. This module inherits from `MusicgenMelodyAttention` as the weights of the module stays
+    untouched. The only required change would be on the forward pass where it needs to correctly call the public API of
+    flash attention and deal with padding tokens in case the input contains any of them.
+    """
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2.__init__
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
+        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
+        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
+        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
+
+    def _reshape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        key_value_states: Optional[torch.Tensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        layer_head_mask: Optional[torch.Tensor] = None,
+        output_attentions: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        # MusicgenMelodyFlashAttention2 attention does not support output_attentions
+        if output_attentions:
+            raise ValueError("MusicgenMelodyFlashAttention2 attention does not support output_attentions")
+
+        # if key_value_states are provided this layer is used as a cross-attention layer
+        # for the decoder
+        is_cross_attention = key_value_states is not None
+
+        bsz, q_len, _ = hidden_states.size()
+
+        # get query proj
+        query_states = self._reshape(self.q_proj(hidden_states), -1, bsz)
+        # get key, value proj
+        # `past_key_value[0].shape[2] == key_value_states.shape[1]`
+        # is checking that the `sequence_length` of the `past_key_value` is the same as
+        # the provided `key_value_states` to support prefix tuning
+        if (
+            is_cross_attention
+            and past_key_value is not None
+            and past_key_value[0].shape[2] == key_value_states.shape[1]
+        ):
+            # reuse k,v, cross_attentions
+            key_states = past_key_value[0].transpose(1, 2)
+            value_states = past_key_value[1].transpose(1, 2)
+        elif is_cross_attention:
+            # cross_attentions
+            key_states = self._reshape(self.k_proj(key_value_states), -1, bsz)
+            value_states = self._reshape(self.v_proj(key_value_states), -1, bsz)
+        elif past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = self._reshape(self.k_proj(hidden_states), -1, bsz)
+            value_states = self._reshape(self.v_proj(hidden_states), -1, bsz)
+            key_states = torch.cat([past_key_value[0].transpose(1, 2), key_states], dim=1)
+            value_states = torch.cat([past_key_value[1].transpose(1, 2), value_states], dim=1)
+        else:
+            # self_attention
+            key_states = self._reshape(self.k_proj(hidden_states), -1, bsz)
+            value_states = self._reshape(self.v_proj(hidden_states), -1, bsz)
+
+        if self.is_decoder:
+            # if cross_attention save Tuple(torch.Tensor, torch.Tensor) of all cross attention key/value_states.
+            # Further calls to cross_attention layer can then reuse all cross-attention
+            # key/value_states (first "if" case)
+            # if uni-directional self-attention (decoder) save Tuple(torch.Tensor, torch.Tensor) of
+            # all previous decoder key/value_states. Further calls to uni-directional self-attention
+            # can concat previous decoder key/value_states to current projected key/value_states (third "elif" case)
+            # if encoder bi-directional self-attention `past_key_value` is always `None`
+            past_key_value = (key_states.transpose(1, 2), value_states.transpose(1, 2))
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value[0].shape[-2]
+
+        # In PEFT, usually we cast the layer norms in float32 for training stability reasons
+        # therefore the input hidden states gets silently casted in float32. Hence, we need
+        # cast them back in the correct dtype just to be sure everything works as expected.
+        # This might slowdown training & inference so it is recommended to not cast the LayerNorms
+        # in fp32. (LlamaRMSNorm handles it correctly)
+
+        input_dtype = query_states.dtype
+        if input_dtype == torch.float32:
+            if torch.is_autocast_enabled():
+                target_dtype = torch.get_autocast_gpu_dtype()
+            # Handle the case where the model is quantized
+            elif hasattr(self.config, "_pre_quantization_dtype"):
+                target_dtype = self.config._pre_quantization_dtype
+            else:
+                target_dtype = self.q_proj.weight.dtype
+
+            logger.warning_once(
+                f"The input hidden states seems to be silently casted in float32, this might be related to"
+                f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
+                f" {target_dtype}."
+            )
+
+            query_states = query_states.to(target_dtype)
+            key_states = key_states.to(target_dtype)
+            value_states = value_states.to(target_dtype)
+
+        attn_output = self._flash_attention_forward(
+            query_states, key_states, value_states, attention_mask, q_len, dropout=self.dropout
+        )
+
+        attn_output = attn_output.reshape(bsz, q_len, -1)
+        attn_output = self.out_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._flash_attention_forward
+    def _flash_attention_forward(
+        self, query_states, key_states, value_states, attention_mask, query_length, dropout=0.0, softmax_scale=None
+    ):
+        """
+        Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
+        first unpad the input, then computes the attention scores and pad the final attention scores.
+        Args:
+            query_states (`torch.Tensor`):
+                Input query states to be passed to Flash Attention API
+            key_states (`torch.Tensor`):
+                Input key states to be passed to Flash Attention API
+            value_states (`torch.Tensor`):
+                Input value states to be passed to Flash Attention API
+            attention_mask (`torch.Tensor`):
+                The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
+                position of padding tokens and 1 for the position of non-padding tokens.
+            dropout (`float`):
+                Attention dropout
+            softmax_scale (`float`, *optional*):
+                The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
+        """
+        if not self._flash_attn_uses_top_left_mask:
+            causal = self.is_causal
+        else:
+            # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
+            causal = self.is_causal and query_length != 1
+
+        # Contains at least one padding token in the sequence
+        if attention_mask is not None:
+            batch_size = query_states.shape[0]
+            query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
+                query_states, key_states, value_states, attention_mask, query_length
+            )
+
+            cu_seqlens_q, cu_seqlens_k = cu_seq_lens
+            max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
+
+            attn_output_unpad = flash_attn_varlen_func(
+                query_states,
+                key_states,
+                value_states,
+                cu_seqlens_q=cu_seqlens_q,
+                cu_seqlens_k=cu_seqlens_k,
+                max_seqlen_q=max_seqlen_in_batch_q,
+                max_seqlen_k=max_seqlen_in_batch_k,
+                dropout_p=dropout,
+                softmax_scale=softmax_scale,
+                causal=causal,
+            )
+
+            attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
+        else:
+            attn_output = flash_attn_func(
+                query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal
+            )
+
+        return attn_output
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._upad_input
+    def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
+        indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
+        batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
+
+        key_layer = index_first_axis(
+            key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
+        )
+        value_layer = index_first_axis(
+            value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
+        )
+        if query_length == kv_seq_len:
+            query_layer = index_first_axis(
+                query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim), indices_k
+            )
+            cu_seqlens_q = cu_seqlens_k
+            max_seqlen_in_batch_q = max_seqlen_in_batch_k
+            indices_q = indices_k
+        elif query_length == 1:
+            max_seqlen_in_batch_q = 1
+            cu_seqlens_q = torch.arange(
+                batch_size + 1, dtype=torch.int32, device=query_layer.device
+            )  # There is a memcpy here, that is very bad.
+            indices_q = cu_seqlens_q[:-1]
+            query_layer = query_layer.squeeze(1)
+        else:
+            # The -q_len: slice assumes left padding.
+            attention_mask = attention_mask[:, -query_length:]
+            query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
+
+        return (
+            query_layer,
+            key_layer,
+            value_layer,
+            indices_q,
+            (cu_seqlens_q, cu_seqlens_k),
+            (max_seqlen_in_batch_q, max_seqlen_in_batch_k),
+        )
+
+
+# Copied from transformers.models.bart.modeling_bart.BartSdpaAttention with Bart->MusicgenMelody
+class MusicgenMelodySdpaAttention(MusicgenMelodyAttention):
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        key_value_states: Optional[torch.Tensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        layer_head_mask: Optional[torch.Tensor] = None,
+        output_attentions: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        """Input shape: Batch x Time x Channel"""
+        if output_attentions or layer_head_mask is not None:
+            # TODO: Improve this warning with e.g. `model.config._attn_implementation = "manual"` once this is implemented.
+            logger.warning_once(
+                "MusicgenMelodyModel is using MusicgenMelodySdpaAttention, but `torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True` or `layer_head_mask` not None. Falling back to the manual attention"
+                ' implementation, but specifying the manual implementation will be required from Transformers version v5.0.0 onwards. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
+            )
+            return super().forward(
+                hidden_states,
+                key_value_states=key_value_states,
+                past_key_value=past_key_value,
+                attention_mask=attention_mask,
+                layer_head_mask=layer_head_mask,
+                output_attentions=output_attentions,
+            )
+
+        # if key_value_states are provided this layer is used as a cross-attention layer
+        # for the decoder
+        is_cross_attention = key_value_states is not None
+
+        bsz, tgt_len, _ = hidden_states.size()
+
+        # get query proj
+        query_states = self.q_proj(hidden_states)
+        # get key, value proj
+        # `past_key_value[0].shape[2] == key_value_states.shape[1]`
+        # is checking that the `sequence_length` of the `past_key_value` is the same as
+        # the provided `key_value_states` to support prefix tuning
+        if (
+            is_cross_attention
+            and past_key_value is not None
+            and past_key_value[0].shape[2] == key_value_states.shape[1]
+        ):
+            # reuse k,v, cross_attentions
+            key_states = past_key_value[0]
+            value_states = past_key_value[1]
+        elif is_cross_attention:
+            # cross_attentions
+            key_states = self._shape(self.k_proj(key_value_states), -1, bsz)
+            value_states = self._shape(self.v_proj(key_value_states), -1, bsz)
+        elif past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
+            value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
+            key_states = torch.cat([past_key_value[0], key_states], dim=2)
+            value_states = torch.cat([past_key_value[1], value_states], dim=2)
+        else:
+            # self_attention
+            key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
+            value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
+
+        if self.is_decoder:
+            # if cross_attention save Tuple(torch.Tensor, torch.Tensor) of all cross attention key/value_states.
+            # Further calls to cross_attention layer can then reuse all cross-attention
+            # key/value_states (first "if" case)
+            # if uni-directional self-attention (decoder) save Tuple(torch.Tensor, torch.Tensor) of
+            # all previous decoder key/value_states. Further calls to uni-directional self-attention
+            # can concat previous decoder key/value_states to current projected key/value_states (third "elif" case)
+            # if encoder bi-directional self-attention `past_key_value` is always `None`
+            past_key_value = (key_states, value_states)
+
+        query_states = self._shape(query_states, tgt_len, bsz)
+
+        # NOTE: SDPA with memory-efficient backend is currently (torch==2.1.2) bugged when using non-contiguous inputs and a custom attn_mask,
+        # but we are fine here as `_shape` do call `.contiguous()`. Reference: https://github.com/pytorch/pytorch/issues/112577
+        attn_output = torch.nn.functional.scaled_dot_product_attention(
+            query_states,
+            key_states,
+            value_states,
+            attn_mask=attention_mask,
+            dropout_p=self.dropout if self.training else 0.0,
+            # The tgt_len > 1 is necessary to match with AttentionMaskConverter.to_causal_4d that does not create a causal mask in case tgt_len == 1.
+            is_causal=self.is_causal and attention_mask is None and tgt_len > 1,
+        )
+
+        if attn_output.size() != (bsz, self.num_heads, tgt_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, tgt_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.transpose(1, 2)
+
+        # Use the `embed_dim` from the config (stored in the class) rather than `hidden_state` because `attn_output` can be
+        # partitioned across GPUs when using tensor-parallelism.
+        attn_output = attn_output.reshape(bsz, tgt_len, self.embed_dim)
+
+        attn_output = self.out_proj(attn_output)
+
+        return attn_output, None, past_key_value
+
+
+MUSICGEN_MELODY_ATTENTION_CLASSES = {
+    "eager": MusicgenMelodyAttention,
+    "sdpa": MusicgenMelodySdpaAttention,
+    "flash_attention_2": MusicgenMelodyFlashAttention2,
+}
+
+
 class MusicgenMelodyDecoderLayer(nn.Module):
     def __init__(self, config: MusicgenMelodyDecoderConfig):
         super().__init__()
         self.embed_dim = config.hidden_size
 
-        self.self_attn = MusicgenMelodyAttention(
+        self.self_attn = MUSICGEN_MELODY_ATTENTION_CLASSES[config._attn_implementation](
             embed_dim=self.embed_dim,
             num_heads=config.num_attention_heads,
             dropout=config.attention_dropout,
             is_decoder=True,
             bias=False,
+            is_causal=True,
+            config=config,
         )
         self.dropout = config.dropout
         self.activation_fn = ACT2FN[config.activation_function]
@@ -414,6 +762,8 @@ class MusicgenMelodyPreTrainedModel(PreTrainedModel):
     base_model_prefix = "model"
     supports_gradient_checkpointing = True
     _no_split_modules = ["MusicgenMelodyDecoderLayer", "MusicgenMelodyAttention"]
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
 
     def _init_weights(self, module):
         std = self.config.initializer_factor
@@ -626,6 +976,7 @@ class MusicgenMelodyDecoder(MusicgenMelodyPreTrainedModel):
 
         self.layers = nn.ModuleList([MusicgenMelodyDecoderLayer(config) for _ in range(config.num_hidden_layers)])
         self.layer_norm = nn.LayerNorm(config.hidden_size)
+        self.attn_implementation = config._attn_implementation
 
         self.gradient_checkpointing = False
         # Initialize weights and apply final processing
@@ -695,9 +1046,21 @@ class MusicgenMelodyDecoder(MusicgenMelodyPreTrainedModel):
 
         input_shape = inputs_embeds.size()[:-1]
 
-        attention_mask = _prepare_4d_causal_attention_mask(
+        if self.attn_implementation == "flash_attention_2":
-            attention_mask, input_shape, inputs_embeds, past_key_values_length
+            attention_mask = attention_mask if (attention_mask is not None and 0 in attention_mask) else None
-        )
+        elif self.attn_implementation == "sdpa" and not output_attentions:
+            # output_attentions=True can not be supported when using SDPA, and we fall back on
+            # the manual implementation that requires a 4D causal mask in all cases.
+            attention_mask = _prepare_4d_causal_attention_mask_for_sdpa(
+                attention_mask,
+                input_shape,
+                inputs_embeds,
+                past_key_values_length,
+            )
+        else:
+            attention_mask = _prepare_4d_causal_attention_mask(
+                attention_mask, input_shape, inputs_embeds, past_key_values_length
+            )
 
         # embed positions
         positions = self.embed_positions(inputs_embeds, past_key_values_length)
@@ -1373,6 +1736,8 @@ class MusicgenMelodyForConditionalGeneration(PreTrainedModel):
     config_class = MusicgenMelodyConfig
     main_input_name = "input_ids"
     supports_gradient_checkpointing = True
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
 
     def __init__(
         self,