modeling_dream.py

# coding=utf-8
# Copyright 2024 The Dream team, HKUNLP Group and the HuggingFace Inc. team. All rights reserved.
#
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
# and OPT and Qwen implementations in this library. It has been modified from its
# original forms to accommodate minor architectural differences compared
# to GPT-NeoX and OPT and Qwen used by the Meta AI and Qwen team that trained the model.
#
# 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.
"""PyTorch Dream model."""



from .modeling_sensevoice import AudioEncoder
from .resampler_projector import ResamplerProjector
import random
from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel

import math
from typing import List, Optional, Tuple, Union
import os
import torch
import torch.utils.checkpoint
from torch import nn

from transformers.activations import ACT2FN
from transformers.cache_utils import Cache, DynamicCache
from transformers.modeling_outputs import (
    # BaseModelOutput,
    MaskedLMOutput,
    ModelOutput
)
from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS
from transformers.modeling_utils import PreTrainedModel
from transformers.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,
)
from transformers import PretrainedConfig
from .configuration_dream import DreamConfig
from .generation_utils import DreamGenerationMixin, DreamGenerationConfig
from  dataclasses import  dataclass
from typing import Any
if is_flash_attn_2_available():
    from transformers.modeling_flash_attention_utils import _flash_attention_forward
from transformers.modeling_outputs import CausalLMOutputWithPast

from .modeling_sensevoice import AudioEncoder
from .resampler_projector import ResamplerProjector

import torch
import random


logger = logging.get_logger(__name__)


@dataclass
class MaskedLMOutput(ModelOutput):
    """
    Base class for masked language models outputs.

    Args:
        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
            Masked language modeling (MLM) loss.
        logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed
        or when `config.output_hidden_states=True`):
                Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has 
                an embedding layer, + one for the output of each layer) of shape `(batch_size, 
                sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed 
        or when `config.output_attentions=True`):
            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    loss: Optional[torch.FloatTensor] = None
    logits: torch.FloatTensor = None
    hidden_states: Optional[Tuple[torch.FloatTensor, ...]] = None
    attentions: Optional[Tuple[torch.FloatTensor, ...]] = None
    past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None

_CHECKPOINT_FOR_DOC = "Dream-7B"
_CONFIG_FOR_DOC = "DreamConfig"
import os
ENFORCE_NUM_ITEMIN_BATCH = os.environ.get("ENFORCE_NUM_ITEMIN_BATCH", False)

@dataclass
class BaseModelOutput(ModelOutput):
    """
    Base class for model's outputs, with potential hidden states and attentions.

    Args:
        last_hidden_state (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
            Sequence of hidden-states at the output of the last layer of the model.
        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed 
        or when `config.output_hidden_states=True`):
            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed 
        or when `config.output_attentions=True`):
            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`.

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
            heads.
    """

    last_hidden_state: torch.FloatTensor = None
    hidden_states: Optional[Tuple[torch.FloatTensor, ...]] = None
    attentions: Optional[Tuple[torch.FloatTensor, ...]] = None
    past_key_values: Optional[Cache] = None
# Copied from transformers.models.llama.modeling_llama.LlamaRMSNorm with Llama->Dream


class DreamRMSNorm(nn.Module):
    def __init__(self, hidden_size, eps=1e-6):
        """
        DreamRMSNorm is equivalent to T5LayerNorm
        """
        super().__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

    def forward(self, hidden_states):
        input_dtype = hidden_states.dtype
        hidden_states = hidden_states.to(torch.float32)
        variance = hidden_states.pow(2).mean(-1, keepdim=True)
        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
        return self.weight * hidden_states.to(input_dtype)

    def extra_repr(self):
        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"


# Copied from transformers.models.llama.modeling_llama.LlamaRotaryEmbedding with Llama->Dream
class DreamRotaryEmbedding(nn.Module):
    def __init__(
        self,
        dim=None,
        max_position_embeddings=2048,
        base=10000,
        device=None,
        scaling_factor=1.0,
        rope_type="default",
        config: Optional[DreamConfig] = None,
    ):
        super().__init__()
        # TODO (joao): remove the `if` below, only used for BC
        self.rope_kwargs = {}
        if config is None:
            logger.warning_once(
                "`DreamRotaryEmbedding` can now be fully parameterized by passing the model config through the "
                "`config` argument. All other arguments will be removed in v4.46"
            )
            self.rope_kwargs = {
                "rope_type": rope_type,
                "factor": scaling_factor,
                "dim": dim,
                "base": base,
                "max_position_embeddings": max_position_embeddings,
            }
            self.rope_type = rope_type
            self.max_seq_len_cached = max_position_embeddings
            self.original_max_seq_len = max_position_embeddings
        else:
            # BC: "rope_type" was originally "type"
            if config.rope_scaling is not None:
                self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
            else:
                self.rope_type = "default"
            self.max_seq_len_cached = config.max_position_embeddings
            self.original_max_seq_len = config.max_position_embeddings

        self.config = config
        self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]

        inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device, **self.rope_kwargs)
        self.register_buffer("inv_freq", inv_freq, persistent=False)
        self.original_inv_freq = self.inv_freq

    def reset_parameters(self):
        inv_freq, self.attention_scaling = self.rope_init_fn(self.config, self.inv_freq.device, **self.rope_kwargs)
        self.register_buffer("inv_freq", inv_freq, persistent=False)
        self.original_inv_freq = self.inv_freq
    

    def _dynamic_frequency_update(self, position_ids, device):
        """
        dynamic RoPE layers should recompute `inv_freq` in the following situations:
        1 - growing beyond the cached sequence length (allow scaling)
        2 - the current sequence length is in the original scale (avoid losing precision with small sequences)
        """
        seq_len = torch.max(position_ids) + 1
        if seq_len > self.max_seq_len_cached:  # growth
            inv_freq, self.attention_scaling = self.rope_init_fn(
                self.config, device, seq_len=seq_len, **self.rope_kwargs
            )
            self.register_buffer("inv_freq", inv_freq, persistent=False)  # TODO joao: may break with compilation
            self.max_seq_len_cached = seq_len

        if seq_len < self.original_max_seq_len and self.max_seq_len_cached > self.original_max_seq_len:  # reset
            self.register_buffer("inv_freq", self.original_inv_freq, persistent=False)
            self.max_seq_len_cached = self.original_max_seq_len

    @torch.no_grad()
    def forward(self, x, position_ids):
        if "dynamic" in self.rope_type:
            self._dynamic_frequency_update(position_ids, device=x.device)

        # Core RoPE block
        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1)
        position_ids_expanded = position_ids[:, None, :].float()
        # Force float32 (see https://github.com/huggingface/transformers/pull/29285)
        device_type = x.device.type
        device_type = device_type if isinstance(device_type, str) and device_type != "mps" else "cpu"
        with torch.autocast(device_type=device_type, enabled=False):
            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
            emb = torch.cat((freqs, freqs), dim=-1)
            cos = emb.cos()
            sin = emb.sin()

        # Advanced RoPE types (e.g. yarn) apply a post-processing scaling factor, equivalent to scaling attention
        cos = cos * self.attention_scaling
        sin = sin * self.attention_scaling

        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)


# Copied from transformers.models.llama.modeling_llama.rotate_half
def rotate_half(x):
    """Rotates half the hidden dims of the input."""
    x1 = x[..., : x.shape[-1] // 2]
    x2 = x[..., x.shape[-1] // 2 :]
    return torch.cat((-x2, x1), dim=-1)


# Copied from transformers.models.llama.modeling_llama.apply_rotary_pos_emb
def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
    """Applies Rotary Position Embedding to the query and key tensors.

    Args:
        q (`torch.Tensor`): The query tensor.
        k (`torch.Tensor`): The key tensor.
        cos (`torch.Tensor`): The cosine part of the rotary embedding.
        sin (`torch.Tensor`): The sine part of the rotary embedding.
        position_ids (`torch.Tensor`, *optional*):
            Deprecated and unused.
        unsqueeze_dim (`int`, *optional*, defaults to 1):
            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
    Returns:
        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
    """
    cos = cos.unsqueeze(unsqueeze_dim)
    sin = sin.unsqueeze(unsqueeze_dim)
    q_embed = (q * cos) + (rotate_half(q) * sin)
    k_embed = (k * cos) + (rotate_half(k) * sin)
    return q_embed, k_embed


# Copied from transformers.models.mistral.modeling_mistral.MistralMLP with Mistral->Dream
class DreamMLP(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.hidden_size = config.hidden_size
        self.intermediate_size = config.intermediate_size
        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
        self.act_fn = ACT2FN[config.hidden_act]

    def forward(self, hidden_state):
        return self.down_proj(self.act_fn(self.gate_proj(hidden_state)) * self.up_proj(hidden_state))


# Copied from transformers.models.llama.modeling_llama.repeat_kv
def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
    """
    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
    """
    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
    if n_rep == 1:
        return hidden_states
    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)


class DreamAttention(nn.Module):
    """
    Multi-headed attention from 'Attention Is All You Need' paper. Modified to use sliding window attention: Longformer
    and "Generating Long Sequences with Sparse Transformers".
    """

    def __init__(self, config: DreamConfig, layer_idx: Optional[int] = None):
        super().__init__()
        self.config = config
        self.layer_idx = layer_idx
        if layer_idx is None:
            logger.warning_once(
                f"Instantiating {self.__class__.__name__} without passing `layer_idx` is not recommended and will "
                "to errors during the forward call, if caching is used. Please make sure to provide a `layer_idx` "
                "when creating this class."
            )

        self.hidden_size = config.hidden_size
        self.num_heads = config.num_attention_heads
        self.head_dim = self.hidden_size // self.num_heads
        self.num_key_value_heads = config.num_key_value_heads
        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
        self.max_position_embeddings = config.max_position_embeddings
        self.rope_theta = config.rope_theta
        self.is_causal = False
        self.attention_dropout = config.attention_dropout

        if (self.head_dim * self.num_heads) != self.hidden_size:
            raise ValueError(
                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
                f" and `num_heads`: {self.num_heads})."
            )
        self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=True)
        self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=True)
        self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=True)
        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)

        self.rotary_emb = DreamRotaryEmbedding(config=self.config)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_value: Optional[Cache] = None,
        output_attentions: bool = False,
        use_cache: bool = False,
        cache_position: Optional[torch.LongTensor] = None,
        position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,  # will become mandatory in v4.46
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
        bsz, q_len, _ = hidden_states.size()

        query_states = self.q_proj(hidden_states)
        key_states = self.k_proj(hidden_states)
        value_states = self.v_proj(hidden_states)

        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)

        if position_embeddings is None:
            logger.warning_once(
                "The attention layers in this model are transitioning from computing the RoPE embeddings internally "
                "through `position_ids` (2D tensor with the indexes of the tokens), to using externally computed "
                "`position_embeddings` (Tuple of tensors, containing cos and sin). In v4.46 `position_ids` will be "
                "removed and `position_embeddings` will be mandatory."
            )
            cos, sin = self.rotary_emb(value_states, position_ids)
        else:
            cos, sin = position_embeddings
        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)

        if past_key_value is not None:
            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}  # Specific to RoPE models
            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)

        # repeat k/v heads if n_kv_heads < n_heads
        key_states = repeat_kv(key_states, self.num_key_value_groups)
        value_states = repeat_kv(value_states, self.num_key_value_groups)

        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
        if attention_mask is not None:  # no matter the length, we just slice it
            causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
            attn_weights = attn_weights + causal_mask

        # upcast attention to fp32
        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
        attn_weights = nn.functional.dropout(attn_weights, p=self.attention_dropout, training=self.training)
        attn_output = torch.matmul(attn_weights, value_states)

        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
            raise ValueError(
                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
                f" {attn_output.size()}"
            )

        attn_output = attn_output.transpose(1, 2).contiguous()
        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)

        attn_output = self.o_proj(attn_output)

        if not output_attentions:
            attn_weights = None

        return attn_output, attn_weights, past_key_value
    

class DreamSdpaAttention(DreamAttention):
    """
    Dream attention module using torch.nn.functional.scaled_dot_product_attention. This module inherits from
    `DreamAttention` as the weights of the module stays untouched. The only changes are on the forward pass to adapt to
    SDPA API.
    """

    # Adapted from DreamAttention.forward
    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_value: Optional[Cache] = None,
        use_cache: bool = False,
        output_attentions: bool = False,
        cache_position: Optional[torch.LongTensor] = None,
        position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,  # will become mandatory in v4.46
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
        if output_attentions:
            # TODO: Improve this warning with e.g. `model.config.attn_implementation = "manual"` 
            # once this is implemented.
            logger.warning_once(
                "DreamModel is using DreamSdpaAttention, but `torch.nn.functional.scaled_dot_product_attention` " \
                " does not support" \
                " `output_attentions=True`. 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=hidden_states,
                attention_mask=attention_mask,
                position_ids=position_ids,
                past_key_value=past_key_value,
                output_attentions=output_attentions,
                use_cache=use_cache,
            )
        bsz, q_len, _ = hidden_states.size()

        query_states = self.q_proj(hidden_states)
        key_states = self.k_proj(hidden_states)
        value_states = self.v_proj(hidden_states)

        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)

        if position_embeddings is None:
            logger.warning_once(
                "The attention layers in this model are transitioning from computing the RoPE embeddings internally "
                "through `position_ids` (2D tensor with the indexes of the tokens), to using externally computed "
                "`position_embeddings` (Tuple of tensors, containing cos and sin). In v4.46 `position_ids` will be "
                "removed and `position_embeddings` will be mandatory."
            )
            cos, sin = self.rotary_emb(value_states, position_ids)
        else:
            cos, sin = position_embeddings
        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)

        if past_key_value is not None:
            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}  # Specific to RoPE models
            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)

        key_states = repeat_kv(key_states, self.num_key_value_groups)
        value_states = repeat_kv(value_states, self.num_key_value_groups)

        # SDPA with memory-efficient backend is currently (torch==2.1.2) bugged with non-contiguous inputs 
        # with custom attn_mask,
        # Reference: https://github.com/pytorch/pytorch/issues/112577.
        if query_states.device.type == "cuda" and attention_mask is not None:
            query_states = query_states.contiguous()
            key_states = key_states.contiguous()
            value_states = value_states.contiguous()

        # We dispatch to SDPA's Flash Attention or Efficient kernels via this `is_causal` if statement 
        # instead of an inline conditional assignment in SDPA to support both torch.compile's dynamic shapes 
        # and full graph options. An inline conditional prevents dynamic shapes from compiling.
        # The q_len > 1 is necessary to match with AttentionMaskConverter.to_causal_4d that does not create 
        # a causal mask in case q_len == 1.

        if attention_mask == None:
            attention_mask = torch.ones([1, 1, 
            query_states.shape[2], query_states.shape[2]]).to(torch.bool).to(query_states.device)
        bool_mask = attention_mask.to(torch.bool)

        attn_output = torch.nn.functional.scaled_dot_product_attention(
            query_states,
            key_states,
            value_states,
            attn_mask=bool_mask ,
            dropout_p=self.attention_dropout if self.training else 0.0,
            is_causal=False, # hard coded
        )

        attn_output = attn_output.transpose(1, 2).contiguous()
        attn_output = attn_output.view(bsz, q_len, self.hidden_size)

        attn_output = self.o_proj(attn_output)

        return attn_output, None, past_key_value

class DreamDecoderLayer(nn.Module):
    def __init__(self, config: DreamConfig, layer_idx: int):
        super().__init__()
        self.hidden_size = config.hidden_size

        if config.sliding_window and config._attn_implementation != "flash_attention_2":
            logger.warning_once(
                f"Sliding Window Attention is enabled but not implemented for "
                f"`{config._attn_implementation}`; "
                "unexpected results may be encountered."
            )
        
        # self.self_attn = Dream_ATTENTION_CLASSES[config._attn_implementation](config, layer_idx)
        self.self_attn = DreamSdpaAttention(config, layer_idx)

        self.mlp = DreamMLP(config)
        self.input_layernorm = DreamRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = DreamRMSNorm(config.hidden_size, eps=config.rms_norm_eps)

    # @torch.compile
    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_value: Optional[Tuple[torch.Tensor]] = None,
        output_attentions: Optional[bool] = False,
        use_cache: Optional[bool] = False,
        cache_position: Optional[torch.LongTensor] = None,
        position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,  
        **kwargs,
    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
        """
        Args:
            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
                `(batch, sequence_length)` where padding elements are indicated by 0.
            output_attentions (`bool`, *optional*):
                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                returned tensors for more detail.
            use_cache (`bool`, *optional*):
                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
                (see `past_key_values`).
            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
            cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*):
                Indices depicting the position of the input sequence tokens in the sequence.
            position_embeddings (`Tuple[torch.FloatTensor, torch.FloatTensor]`, *optional*):
                Tuple containing the cosine and sine positional embeddings of shape `(batch_size, seq_len, head_dim)`,
                with `head_dim` being the embedding dimension of each attention head.
            kwargs (`dict`, *optional*):
                Arbitrary kwargs to be ignored, used for FSDP and other methods that injects code
                into the model
        """

        residual = hidden_states

        hidden_states = self.input_layernorm(hidden_states)

        hidden_states, self_attn_weights, present_key_value = self.self_attn(
            hidden_states=hidden_states,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_value=past_key_value,
            output_attentions=output_attentions,
            use_cache=use_cache,
            cache_position=cache_position,
            position_embeddings=position_embeddings,
        )
        hidden_states = residual + hidden_states

        # Fully Connected
        residual = hidden_states
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = residual + hidden_states

        outputs = (hidden_states,)

        if output_attentions:
            outputs += (self_attn_weights,)

        if use_cache:
            outputs += (present_key_value,)

        return outputs

class DreamPreTrainedModel(PreTrainedModel):
    config_class = DreamConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["DreamDecoderLayer"]
    _skip_keys_device_placement = "past_key_values"
    _supports_flash_attn_2 = True
    _supports_sdpa = True
    _supports_cache_class = True
    _supports_quantized_cache = True
    _supports_static_cache = True

    def _init_weights(self, module):
        std = self.config.initializer_range
        if isinstance(module, nn.Linear):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.padding_idx is not None:
                module.weight.data[module.padding_idx].zero_()

    @classmethod
    def from_pretrained(
        cls,
        pretrained_model_name_or_path: Optional[Union[str, os.PathLike]],
        *model_args,
        config: Optional[Union[PretrainedConfig, str, os.PathLike]] = None,
        cache_dir: Optional[Union[str, os.PathLike]] = None,
        ignore_mismatched_sizes: bool = False,
        force_download: bool = False,
        local_files_only: bool = False,
        token: Optional[Union[str, bool]] = None,
        revision: str = "main",
        use_safetensors: Optional[bool] = None,
        weights_only: bool = True,
        **kwargs,
    ):
        _ = None
        try:
            _model,_  = super().from_pretrained(
                pretrained_model_name_or_path,
                *model_args,
                config=config,
                cache_dir=cache_dir,
                ignore_mismatched_sizes=ignore_mismatched_sizes,
                force_download=force_download,
                local_files_only=local_files_only,
                token=token,
                revision=revision,
                use_safetensors=use_safetensors,
                weights_only=weights_only,
                **kwargs,
            )
        except Exception as e:
            _model = super().from_pretrained(
                pretrained_model_name_or_path,
                *model_args,
                config=config,
                cache_dir=cache_dir,
                ignore_mismatched_sizes=ignore_mismatched_sizes,
                force_download=force_download,
                local_files_only=local_files_only,
                token=token,
                revision=revision,
                use_safetensors=use_safetensors,
                weights_only=weights_only,
                **kwargs,
            )

        resume_download = kwargs.get("resume_download", None)
        proxies = kwargs.get("proxies", None)
        subfolder = kwargs.get("subfolder", "")
        from_auto_class = kwargs.get("_from_auto", False)
        from_pipeline = kwargs.get("_from_pipeline", None)
        _model.generation_config= DreamGenerationConfig.from_pretrained(
            pretrained_model_name_or_path,
            cache_dir=cache_dir,
            force_download=force_download,
            resume_download=resume_download,
            proxies=proxies,
            local_files_only=local_files_only,
            token=token,
            revision=revision,
            subfolder=subfolder,
            _from_auto=from_auto_class,
            _from_pipeline=from_pipeline,
        )
        if _ is not None:
            return _model,_
        return _model
    
class DreamPrefixLMCache(Cache):
    
    def __init__(self):
        super().__init__()
        self.past_key_values = {}
        # this will not be updated beyond the prefilling phase
        
    def update(
        self,
        key_states: torch.Tensor,
        value_states: torch.Tensor,
        layer_idx: int,
        cache_kwargs = None,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        if layer_idx in self.past_key_values:
            past_key, past_value = self.past_key_values[layer_idx]
            key_states = torch.cat((past_key, key_states), dim=-2)
            value_states = torch.cat((past_value, value_states), dim=-2)
            return key_states,value_states
        else:
            self.past_key_values[layer_idx] = (key_states, value_states)
            return key_states, value_states

    def get_seq_length(self, layer_idx: Optional[int] = 0) -> int:
        """Returns the sequence length of the cached states. A layer index can be optionally passed."""
        # TODO: deprecate this function in favor of `cache_position`
        if len(self.past_key_values) == 0:
            return 0
        else:
            return self.past_key_values[0][0].shape[-2]
        
    def get_max_cache_shape(self) -> Optional[int]:
        return None
 

import deepspeed
class DreamBaseModel(DreamPreTrainedModel):#
    """
    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`DreamDecoderLayer`]

    Args:
        config: DreamConfig
    """

    def __init__(self, config: DreamConfig):
        super().__init__(config)
        self.padding_idx = config.pad_token_id
        self.vocab_size = config.vocab_size

        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
        self.layers = nn.ModuleList(
            [DreamDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
        )
        self._attn_implementation = config._attn_implementation
        self.norm = DreamRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.rotary_emb = DreamRotaryEmbedding(config=config)

        self.gradient_checkpointing = False
        # Initialize weights and apply final processing

        self.audio_model = AudioEncoder()
        self.audio_projection = ResamplerProjector(512, config.hidden_size)

        self.post_init()

    def get_input_embeddings(self):
        return self.embed_tokens

    def set_input_embeddings(self, value):
        self.embed_tokens = value

    def forward(
        self,
        input_ids: torch.LongTensor = None,
        attention_mask: Optional[torch.Tensor] = None,
        audios: Optional[torch.FloatTensor] = None,
        audio_indices: Optional[torch.LongTensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[List[torch.FloatTensor]] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        cache_position: Optional[torch.LongTensor] = None,
    ) -> Union[Tuple, BaseModelOutput]:

        if (past_key_values is None or len(past_key_values) == 0) and audios is not None:
            audio_embeds, audio_lengths = self.audio_model(audios)
            assert audio_embeds.shape[0] == len(audios)
            fake_audios = None

            audio_embeds = self.audio_projection(audio_embeds)

        elif self.training:
            device = self.get_input_embeddings().weight.data.device
            dtype = self.get_input_embeddings().weight.data.dtype
            fake_audios = torch.ones((1, 1, 560), dtype=dtype, device=device)
            audio_embeds, audio_lengths = self.audio_model(fake_audios)
            audio_embeds = self.audio_projection(audio_embeds)

        else:
            fake_audios = None
            audio_embeds = None

    
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        use_cache = use_cache if use_cache is not None else self.config.use_cache

        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if (input_ids is None) ^ (inputs_embeds is not None):
            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")

        if self.gradient_checkpointing and self.training:
            if use_cache:
                logger.warning_once(
                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                )
                use_cache = False

        if inputs_embeds is None:
            inputs_embeds = self.embed_tokens(input_ids)
        


        if fake_audios is not None:
            inputs_embeds = inputs_embeds + audio_embeds.mean() * 0.0
        elif audio_embeds is not None:
            inputs_embeds = inputs_embeds.clone()
            for audio_embeds_, audio_lengths_, audio_indices_ in zip(audio_embeds, audio_lengths, audio_indices,):
                # print(f"{audio_embeds_.size()=} {audio_lengths_=} {audio_indices_.size()=}")
                audio_embeds_ = audio_embeds_[:audio_lengths_, ...]
                audio_embeds_ = audio_embeds_.to(inputs_embeds.device)
                indices_b, indices_s = audio_indices_.to(inputs_embeds.device).unbind(dim=0)
                inputs_embeds[indices_b.view(-1), indices_s.view(-1)] = audio_embeds_.view(-1, audio_embeds_.shape[-1])


        if use_cache and past_key_values is None:
            past_key_values = DreamPrefixLMCache()

        if cache_position is None:
            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
            cache_position = torch.arange(
                past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
            )

        if position_ids is None:
            position_ids = cache_position.unsqueeze(0)

        hidden_states = inputs_embeds

        # create position embeddings to be shared across the decoder layers
        position_embeddings = self.rotary_emb(hidden_states, position_ids)

        # decoder layers
        all_hidden_states = () if output_hidden_states else None
        all_self_attns = () if output_attentions else None

        for decoder_layer in self.layers:
            if output_hidden_states:
                all_hidden_states += (hidden_states,)

            if self.gradient_checkpointing and self.training:
                layer_outputs = deepspeed.checkpointing.checkpoint(
                    decoder_layer,
                    hidden_states,
                    attention_mask,
                    position_ids,
                    past_key_values,
                    output_attentions,
                    use_cache,
                    cache_position,
                    position_embeddings,
                )
            else:
                layer_outputs = decoder_layer(
                    hidden_states,
                    attention_mask=attention_mask,
                    position_ids=position_ids,
                    past_key_value=past_key_values,
                    output_attentions=output_attentions,
                    use_cache=use_cache,
                    cache_position=cache_position,
                    position_embeddings=position_embeddings,
                )

            # breakpoint()
            if isinstance(layer_outputs,torch.Tensor):
                layer_outputs = (layer_outputs,None)
            hidden_states = layer_outputs[0]

            if output_attentions:
                all_self_attns += (layer_outputs[1],)

        hidden_states = self.norm(hidden_states)

        # add hidden states from the last decoder layer
        if output_hidden_states:
            all_hidden_states += (hidden_states,)

        if not return_dict:
            return tuple(v for v in [hidden_states, all_hidden_states, all_self_attns] if v is not None)
        return BaseModelOutput(
            last_hidden_state=hidden_states,
            hidden_states=all_hidden_states,
            attentions=all_self_attns,
            past_key_values=past_key_values,
        )


class DreamModel(DreamGenerationMixin, DreamPreTrainedModel):
    _tied_weights_keys = ["lm_head.weight"]

    def __init__(self, config):
        super().__init__(config)
        self.model = DreamBaseModel(config)
        self.vocab_size = config.vocab_size
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

        # Initialize weights and apply final processing
        self.tokenizer = None
        self.post_init()

    def reset_rope_parameters(self):
        self.model.rotary_emb.reset_parameters()
        for layer in self.model.layers:
            layer.self_attn.rotary_emb.reset_parameters()

    def get_input_embeddings(self):
        return self.model.embed_tokens

    def set_input_embeddings(self, value):
        self.model.embed_tokens = value

    def get_output_embeddings(self):
        return self.lm_head

    def set_output_embeddings(self, new_embeddings):
        self.lm_head = new_embeddings

    def set_decoder(self, decoder):
        self.model = decoder

    def get_decoder(self):
        return self.model

    def forward(
        self,
        input_ids: torch.LongTensor = None,
        attention_mask: Optional[torch.Tensor] = None,
        audios: Optional[torch.FloatTensor] = None,
        audio_indices: Optional[torch.LongTensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[List[torch.FloatTensor]] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        cache_position: Optional[torch.LongTensor] = None,
        num_logits_to_keep: int = 0,
        num_items_in_batch: int = None,
        **loss_kwargs,
    ) -> Union[Tuple, MaskedLMOutput]:

        num_items_in_batch = None
        if ENFORCE_NUM_ITEMIN_BATCH:
            num_items_in_batch = labels.ne(-100).sum()
            num_items_in_batch = torch.distributed.reduce(num_items_in_batch)

        is_new = position_ids == 0
        # is_new[0] = True
        segment_id = torch.cumsum(is_new.long(), dim=1) - 1
        new_attention_mask = (segment_id.unsqueeze(1) == segment_id.unsqueeze(2)).long()
        mask = attention_mask.unsqueeze(-1)  # [bs, len, 1]
        new_attention_mask = new_attention_mask * mask 

        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.model(
            input_ids=input_ids,
            attention_mask=new_attention_mask,
            audios=audios,
            audio_indices=audio_indices,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            cache_position=cache_position,
        )
        hidden_states = outputs[0]
        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
        logits = self.lm_head(hidden_states[:, -num_logits_to_keep:, :])

        loss = None
        if labels is not None:
            if ENFORCE_NUM_ITEMIN_BATCH:
                assert num_items_in_batch is not None, \
                    "num_items_in_batch must be provided if ENFORCE_NUM_ITEMIN_BATCH is True"
            loss = self.loss_function(logits, labels, 
                                      self.vocab_size,num_items_in_batch=num_items_in_batch, **loss_kwargs)

        if not return_dict:
            output = (logits,) + outputs[1:]
            return (loss,) + output if loss is not None else output

        return MaskedLMOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
            past_key_values=outputs.past_key_values
        )
        
    def forward_dream(
        self,
        input_ids: torch.LongTensor = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[List[torch.FloatTensor]] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        cache_position: Optional[torch.LongTensor] = None,
        num_logits_to_keep: int = 0,
        **loss_kwargs,
    ) -> Union[Tuple, MaskedLMOutput]:
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
        attention_mask = None
        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
        outputs = self.model(
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            cache_position=cache_position,
        )

        hidden_states = outputs[0]
        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
        logits = self.lm_head(hidden_states[:, -num_logits_to_keep:, :])

        loss = None
        if labels is not None:
            loss = self.loss_function(logits, labels, self.vocab_size, **loss_kwargs)

        if not return_dict:
            output = (logits,) + outputs[1:]
            return (loss,) + output if loss is not None else output

        return MaskedLMOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
            past_key_values=outputs.past_key_values,
        )

    @torch.no_grad()
    def generate(
        self,
        input_ids: Optional[torch.Tensor] = None,
        audios: Optional[torch.FloatTensor] = None,
        audio_indices: Optional[torch.LongTensor] = None,
        max_new_tokens=512,
        steps=512,
        temperature=0.2,
        top_p=0.95,
        alg_temp=0.,
        alg="entropy",
        output_history=False,
        **kwargs,
    ):
        # modalities = kwargs.pop("modalities", None) if "modalities" in kwargs and modalities is None else modalities
        position_ids = kwargs.pop("position_ids", None)
        attention_mask = kwargs.pop("attention_mask", None)
        if "inputs_embeds" in kwargs:
            raise NotImplementedError("`inputs_embeds` is not supported")

        if audios is not None:
            audio_embeds, audio_lengths = self.model.audio_model(audios)
            assert audio_embeds.shape[0] == len(audios)
            audio_embeds = self.model.audio_projection(audio_embeds)
        else:
            audio_embeds = None

        inputs_embeds = self.model.embed_tokens(input_ids)
        
        if audio_embeds is not None:
            inputs_embeds = inputs_embeds.clone()
            for audio_embeds_, audio_lengths_, audio_indices_ in zip(audio_embeds, audio_lengths, audio_indices,):
                # print(f"{audio_embeds_.size()=} {audio_lengths_=} {audio_indices_.size()=}")
                audio_embeds_ = audio_embeds_[:audio_lengths_, ...]
                audio_embeds_ = audio_embeds_.to(inputs_embeds.device)
                indices_b, indices_s = audio_indices_.to(inputs_embeds.device).unbind(dim=0)
                inputs_embeds[indices_b.view(-1), indices_s.view(-1)] = audio_embeds_.view(-1, audio_embeds_.shape[-1])

        return self.diffusion_generate(
            None,
            inputs_embeds=inputs_embeds,
            max_new_tokens=max_new_tokens,
            output_history=output_history,
            return_dict_in_generate=True,
            steps=steps,
            temperature=temperature,
            top_p=top_p,
            alg=alg,
            alg_temp=alg_temp,
            **kwargs
        )