modeling_typhoonaudio.py

"""
Adapted from ByteDance's SALMONN (https://github.com/bytedance/SALMONN). 
Please follow the original copyright of the SALMONN project. 
"""

# ---------------------------------------------------- # 
import torch
from torch import Tensor, device, dtype, nn
import torch.nn.functional as F
import random
import numpy as np
from peft import LoraConfig, TaskType, get_peft_model
from transformers import (
    WhisperFeatureExtractor,
    WhisperModel,
    PreTrainedModel,
    AutoTokenizer, 
    AutoModelForCausalLM
)
import soundfile as sf
import librosa
from .configuration_typhoonaudio import TyphoonAudioConfig
# ---------------------------------------------------- # 
# QFormer: https://github.com/huggingface/transformers/blob/v4.15.0/src/transformers/models/bert
import math
import os
import warnings
from dataclasses import dataclass
from typing import Optional, Tuple, Dict, Any, Union
import torch.utils.checkpoint
from torch.nn import CrossEntropyLoss
from transformers.activations import ACT2FN
from transformers.file_utils import (
    ModelOutput,
)
from transformers.modeling_outputs import (
    BaseModelOutputWithPastAndCrossAttentions,
    BaseModelOutputWithPoolingAndCrossAttentions,
    CausalLMOutputWithCrossAttentions,
    MaskedLMOutput,
    MultipleChoiceModelOutput,
    NextSentencePredictorOutput,
    QuestionAnsweringModelOutput,
    SequenceClassifierOutput,
    TokenClassifierOutput,
)
from transformers.modeling_utils import (
    PreTrainedModel,
    apply_chunking_to_forward,
    find_pruneable_heads_and_indices,
    prune_linear_layer,
)
from transformers.models.bert.configuration_bert import BertConfig
# ---------------------------------------------------------- # 
# BEATs:  https://github.com/microsoft/unilm/tree/master/beats
from torch.nn import LayerNorm, Parameter
import torch.distributed as distributed
import torchaudio.compliance.kaldi as ta_kaldi
import logging
try:
    from einops import rearrange, repeat
except ImportError:
    pass

logger = logging.getLogger(__name__)

# ---------------------------------------------------------- # 

class TyphoonAudio(PreTrainedModel):
    config_class = TyphoonAudioConfig

    def __init__(self, config):
        super().__init__(config)

        # Can we fix this to not be manual?
        if config.dtype == "float16":
            self.torch_dtype = torch.float16
        elif config.dtype == "bfloat16":
            self.torch_dtype = torch.bfloat16
        elif config.dtype == "float32":
            self.torch_dtype = torch.float32
        else:
            raise ValueError("dtype not supported")

        # feature_extractor
        self.feature_extractor = WhisperFeatureExtractor.from_pretrained(config.whisper_path)

        # whisper encoder
        self.speech_encoder = WhisperModel.from_pretrained(
            config.whisper_path, 
            torch_dtype=self.torch_dtype
        ).encoder
        self.ln_speech = nn.LayerNorm(self.speech_encoder.config.d_model, dtype=self.torch_dtype)

        # beats
        beats_cfg = BEATsConfig()
        beats = BEATs(beats_cfg)
        self.beats = beats
        self.ln_audio = nn.LayerNorm(self.beats.cfg.encoder_embed_dim, dtype=self.torch_dtype)
        for name, param in self.beats.named_parameters():
            param.requires_grad = False
        self.beats.eval()
        self.beats.to(self.torch_dtype)

        # init speech Qformer
        self.speech_Qformer, self.speech_query_tokens = self.init_speech_Qformer(
            config.speech_qformer_token_num,
            self.speech_encoder.config.d_model + self.beats.cfg.encoder_embed_dim,
            config.speech_qformer_layer,
            torch_dtype=self.torch_dtype
        )
        self.second_per_frame = config.second_per_frame
        self.second_stride = config.second_stride

        # llm - llama3
        self.llama_model = AutoModelForCausalLM.from_pretrained(
            config.llm_path,
            torch_dtype=self.torch_dtype, # should it be torch.bfloat16? 
        )

        # lora
        self.lora = config.lora
        if self.lora:
            # target_modules = None
            self.peft_config = LoraConfig(
                task_type=TaskType.CAUSAL_LM, 
                inference_mode=True, 
                r=config.lora_rank, 
                lora_alpha=config.lora_alpha, 
                lora_dropout=config.lora_dropout,
                # target_modules=target_modules,
            )
            self.llama_model = get_peft_model(self.llama_model, self.peft_config)

        # tokenizer
        self.llama_tokenizer = AutoTokenizer.from_pretrained(config.llm_path, use_fast=False)
        self.llama_tokenizer.add_special_tokens({'pad_token': '[PAD]'}) 
        self.llama_tokenizer.padding_side = "right"

        # proj
        self.speech_llama_proj = nn.Linear(
            self.speech_Qformer.config.hidden_size, self.llama_model.config.hidden_size,
            dtype=self.torch_dtype
        )

    def load_beats(self, beats_path):
        beats_checkpoint = torch.load(beats_path, map_location='cpu')
        self.beats.load_state_dict(beats_checkpoint['model'])

    def load_adapter(self, ckpt_path):
        ckpt_dict = torch.load(ckpt_path)['model']
        self.load_state_dict(ckpt_dict, strict=False)

    def forward(self, **kwargs):
        raise Exception("Direct forward pass is not supported. For training, please refer to the training recipe of Typhoon-Audio.")

    def generate(
        self,
        wav_path,
        prompt,
        prompt_pattern,
        device='cuda:0',
        max_length=150,
        num_beams=4,
        do_sample=True,
        min_length=1,
        top_p=0.9,
        repetition_penalty=1.0,
        length_penalty=1.0,
        temperature=1.0,
        streamer=None
    ):
        # read wav
        wav, sr = sf.read(wav_path)
        if len(wav.shape) == 2:
            wav = wav[:, 0]
        if len(wav) > 30 * sr:
            wav = wav[: 30 * sr]
        if sr != 16000:
            wav = librosa.resample(wav, orig_sr=sr, target_sr=16000, res_type="fft")
        
        # whisper
        spectrogram = self.feature_extractor(wav, return_tensors="pt", sampling_rate=16000).input_features.to(device).to(self.torch_dtype) # [1, 80, 3000]
        speech_embeds = self.speech_encoder(spectrogram, return_dict=True).last_hidden_state

        # beats
        raw_wav = torch.from_numpy(wav).to(device).unsqueeze(0)
        audio_padding_mask = torch.zeros(raw_wav.shape, device=device).bool()
        audio_embeds, _ = self.beats.extract_features(raw_wav, padding_mask=audio_padding_mask, feature_only=True, torch_dtype=self.torch_dtype)

        # auditory embeds
        speech_embeds = self.ln_speech(speech_embeds)
        audio_embeds = self.ln_audio(audio_embeds)
        audio_embeds = F.pad(audio_embeds, (0, 0, 0, speech_embeds.size(1) - audio_embeds.size(1)))
        speech_embeds = torch.cat([speech_embeds, audio_embeds], dim=-1)

        # split frames
        B, T, C = speech_embeds.shape
        kernel = round(T * self.second_per_frame / 30.0)
        stride = round(T * self.second_stride / 30.0)
        kernel = (1, kernel)
        stride = (1, stride)
        speech_embeds_tr = speech_embeds.transpose(1, 2).unsqueeze(2)
        speech_embeds_overlap = F.unfold(speech_embeds_tr, kernel_size=kernel, dilation=1, padding=0, stride=stride)
        _, _, L = speech_embeds_overlap.shape
        speech_embeds_overlap = speech_embeds_overlap.view(B, -1, kernel[1], L)
        speech_embeds_overlap = torch.permute(speech_embeds_overlap, [0, 3, 2, 1])
        speech_embeds = speech_embeds_overlap.reshape(-1, kernel[1], C)
        speech_atts = torch.ones(speech_embeds.size()[:-1], dtype=torch.long, device=speech_embeds.device)

        # Qformer
        query_tokens = self.speech_query_tokens.expand(speech_embeds.shape[0], -1, -1)
        query_output = self.speech_Qformer.bert(
            query_embeds=query_tokens,
            encoder_hidden_states=speech_embeds,
            encoder_attention_mask=speech_atts,
            return_dict=True,
        )
        speech_embeds = self.speech_llama_proj(query_output.last_hidden_state)
        speech_embeds = speech_embeds.view(B, -1, speech_embeds.size(2)).contiguous()
        speech_atts = torch.ones(speech_embeds.size()[:-1], dtype=torch.long).to(speech_embeds.device)

        # "<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\n<Speech><SpeechHere></Speech> {}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
        embed_tokens = self.llama_model.model.model.embed_tokens if self.lora else self.llama_model.model.embed_tokens
        prompt_left, prompts_right = prompt_pattern.format(prompt).split('<SpeechHere>')
        prompt_left_ids = self.llama_tokenizer(
            prompt_left,
            return_tensors="pt",
            add_special_tokens=False
        ).to(speech_embeds.device).input_ids
        prompt_left_embeds = embed_tokens(prompt_left_ids)
        prompt_right_ids = self.llama_tokenizer(
            prompts_right,
            return_tensors="pt",
            add_special_tokens=False
        ).to(speech_embeds.device).input_ids
        prompt_right_embeds = embed_tokens(prompt_right_ids)

        bos_embeds = self.llama_model.model.embed_tokens(
            torch.ones(
                [1, 1],
                dtype=torch.long,
                device=device,
            ) * self.llama_tokenizer.bos_token_id
        ) if not self.lora else self.llama_model.model.model.embed_tokens(
            torch.ones(
                [1, 1],
                dtype=torch.long,
                device=device,
            ) * self.llama_tokenizer.bos_token_id
        )

        embeds = torch.cat([bos_embeds, prompt_left_embeds, speech_embeds, prompt_right_embeds], dim=1)
        atts = torch.ones(embeds.size()[:-1], dtype=torch.long).to(embeds.device)

        # generate
        output = self.llama_model.generate(
            inputs_embeds=embeds,
            max_length=max_length,
            num_beams=num_beams,
            do_sample=do_sample,
            min_length=min_length,
            top_p=top_p,
            repetition_penalty=repetition_penalty,
            length_penalty=length_penalty,
            temperature=temperature,
            attention_mask=atts,
            bos_token_id=self.llama_tokenizer.bos_token_id,
            eos_token_id=self.llama_tokenizer.eos_token_id,
            pad_token_id=self.llama_tokenizer.pad_token_id,
            streamer=streamer,
        )
        output_text = self.llama_tokenizer.batch_decode(output, add_special_tokens=False, skip_special_tokens=True)
        return output_text[0]

    def init_speech_Qformer(self, num_query_token, speech_width, num_hidden_layers=2, torch_dtype="float16"):
        encoder_config = BertConfig()
        encoder_config.num_hidden_layers = num_hidden_layers
        encoder_config.encoder_width = speech_width
        encoder_config.add_cross_attention = True
        encoder_config.cross_attention_freq = 1
        encoder_config.query_length = num_query_token
        Qformer = BertLMHeadModel(config=encoder_config)
        Qformer.to(torch_dtype)
        query_tokens = nn.Parameter(
            torch.zeros(1, num_query_token, encoder_config.hidden_size, dtype=torch_dtype),
        )
        query_tokens.data.normal_(mean=0.0, std=encoder_config.initializer_range)
        return Qformer, query_tokens

class BertEmbeddings(nn.Module):
    """Construct the embeddings from word and position embeddings."""

    def __init__(self, config):
        super().__init__()
        self.word_embeddings = nn.Embedding(
            config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id
        )
        self.position_embeddings = nn.Embedding(
            config.max_position_embeddings, config.hidden_size
        )

        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
        # any TensorFlow checkpoint file
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
        self.register_buffer(
            "position_ids", torch.arange(config.max_position_embeddings).expand((1, -1))
        )
        self.position_embedding_type = getattr(
            config, "position_embedding_type", "absolute"
        )

        self.config = config

    def forward(
        self,
        input_ids=None,
        position_ids=None,
        query_embeds=None,
        past_key_values_length=0,
    ):
        if input_ids is not None:
            seq_length = input_ids.size()[1]
        else:
            seq_length = 0

        if position_ids is None:
            position_ids = self.position_ids[
                :, past_key_values_length : seq_length + past_key_values_length
            ].clone()

        if input_ids is not None:
            embeddings = self.word_embeddings(input_ids)
            if self.position_embedding_type == "absolute":
                position_embeddings = self.position_embeddings(position_ids)
                embeddings = embeddings + position_embeddings

            if query_embeds is not None:
                embeddings = torch.cat((query_embeds, embeddings), dim=1)
        else:
            embeddings = query_embeds

        embeddings = self.LayerNorm(embeddings)
        embeddings = self.dropout(embeddings)
        return embeddings


class BertSelfAttention(nn.Module):
    def __init__(self, config, is_cross_attention):
        super().__init__()
        self.config = config
        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(
            config, "embedding_size"
        ):
            raise ValueError(
                "The hidden size (%d) is not a multiple of the number of attention "
                "heads (%d)" % (config.hidden_size, config.num_attention_heads)
            )

        self.num_attention_heads = config.num_attention_heads
        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
        self.all_head_size = self.num_attention_heads * self.attention_head_size

        self.query = nn.Linear(config.hidden_size, self.all_head_size)
        if is_cross_attention:
            self.key = nn.Linear(config.encoder_width, self.all_head_size)
            self.value = nn.Linear(config.encoder_width, self.all_head_size)
        else:
            self.key = nn.Linear(config.hidden_size, self.all_head_size)
            self.value = nn.Linear(config.hidden_size, self.all_head_size)

        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
        self.position_embedding_type = getattr(
            config, "position_embedding_type", "absolute"
        )
        if (
            self.position_embedding_type == "relative_key"
            or self.position_embedding_type == "relative_key_query"
        ):
            self.max_position_embeddings = config.max_position_embeddings
            self.distance_embedding = nn.Embedding(
                2 * config.max_position_embeddings - 1, self.attention_head_size
            )
        self.save_attention = False

    def save_attn_gradients(self, attn_gradients):
        self.attn_gradients = attn_gradients

    def get_attn_gradients(self):
        return self.attn_gradients

    def save_attention_map(self, attention_map):
        self.attention_map = attention_map

    def get_attention_map(self):
        return self.attention_map

    def transpose_for_scores(self, x):
        new_x_shape = x.size()[:-1] + (
            self.num_attention_heads,
            self.attention_head_size,
        )
        x = x.view(*new_x_shape)
        return x.permute(0, 2, 1, 3)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_value=None,
        output_attentions=False,
    ):

        # If this is instantiated as a cross-attention module, the keys
        # and values come from an encoder; the attention mask needs to be
        # such that the encoder's padding tokens are not attended to.
        is_cross_attention = encoder_hidden_states is not None

        if is_cross_attention:
            key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
            value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
            attention_mask = encoder_attention_mask
        elif past_key_value is not None:
            key_layer = self.transpose_for_scores(self.key(hidden_states))
            value_layer = self.transpose_for_scores(self.value(hidden_states))
            key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
            value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
        else:
            key_layer = self.transpose_for_scores(self.key(hidden_states))
            value_layer = self.transpose_for_scores(self.value(hidden_states))

        mixed_query_layer = self.query(hidden_states)

        query_layer = self.transpose_for_scores(mixed_query_layer)

        past_key_value = (key_layer, value_layer)

        # Take the dot product between "query" and "key" to get the raw attention scores.
        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))

        if (
            self.position_embedding_type == "relative_key"
            or self.position_embedding_type == "relative_key_query"
        ):
            seq_length = hidden_states.size()[1]
            position_ids_l = torch.arange(
                seq_length, dtype=torch.long, device=hidden_states.device
            ).view(-1, 1)
            position_ids_r = torch.arange(
                seq_length, dtype=torch.long, device=hidden_states.device
            ).view(1, -1)
            distance = position_ids_l - position_ids_r
            positional_embedding = self.distance_embedding(
                distance + self.max_position_embeddings - 1
            )
            positional_embedding = positional_embedding.to(
                dtype=query_layer.dtype
            )  # fp16 compatibility

            if self.position_embedding_type == "relative_key":
                relative_position_scores = torch.einsum(
                    "bhld,lrd->bhlr", query_layer, positional_embedding
                )
                attention_scores = attention_scores + relative_position_scores
            elif self.position_embedding_type == "relative_key_query":
                relative_position_scores_query = torch.einsum(
                    "bhld,lrd->bhlr", query_layer, positional_embedding
                )
                relative_position_scores_key = torch.einsum(
                    "bhrd,lrd->bhlr", key_layer, positional_embedding
                )
                attention_scores = (
                    attention_scores
                    + relative_position_scores_query
                    + relative_position_scores_key
                )

        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
        if attention_mask is not None:
            # Apply the attention mask is (precomputed for all layers in BertModel forward() function)
            attention_scores = attention_scores + attention_mask

        # Normalize the attention scores to probabilities.
        attention_probs = nn.Softmax(dim=-1)(attention_scores)

        if is_cross_attention and self.save_attention:
            self.save_attention_map(attention_probs)
            attention_probs.register_hook(self.save_attn_gradients)

        # This is actually dropping out entire tokens to attend to, which might
        # seem a bit unusual, but is taken from the original Transformer paper.
        attention_probs_dropped = self.dropout(attention_probs)

        # Mask heads if we want to
        if head_mask is not None:
            attention_probs_dropped = attention_probs_dropped * head_mask

        context_layer = torch.matmul(attention_probs_dropped, value_layer)

        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
        context_layer = context_layer.view(*new_context_layer_shape)

        outputs = (
            (context_layer, attention_probs) if output_attentions else (context_layer,)
        )

        outputs = outputs + (past_key_value,)
        return outputs


class BertSelfOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states, input_tensor):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


class BertAttention(nn.Module):
    def __init__(self, config, is_cross_attention=False):
        super().__init__()
        self.self = BertSelfAttention(config, is_cross_attention)
        self.output = BertSelfOutput(config)
        self.pruned_heads = set()

    def prune_heads(self, heads):
        if len(heads) == 0:
            return
        heads, index = find_pruneable_heads_and_indices(
            heads,
            self.self.num_attention_heads,
            self.self.attention_head_size,
            self.pruned_heads,
        )

        # Prune linear layers
        self.self.query = prune_linear_layer(self.self.query, index)
        self.self.key = prune_linear_layer(self.self.key, index)
        self.self.value = prune_linear_layer(self.self.value, index)
        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)

        # Update hyper params and store pruned heads
        self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
        self.self.all_head_size = (
            self.self.attention_head_size * self.self.num_attention_heads
        )
        self.pruned_heads = self.pruned_heads.union(heads)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_value=None,
        output_attentions=False,
    ):
        self_outputs = self.self(
            hidden_states,
            attention_mask,
            head_mask,
            encoder_hidden_states,
            encoder_attention_mask,
            past_key_value,
            output_attentions,
        )
        attention_output = self.output(self_outputs[0], hidden_states)

        outputs = (attention_output,) + self_outputs[
            1:
        ]  # add attentions if we output them
        return outputs


class BertIntermediate(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
        if isinstance(config.hidden_act, str):
            self.intermediate_act_fn = ACT2FN[config.hidden_act]
        else:
            self.intermediate_act_fn = config.hidden_act

    def forward(self, hidden_states):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.intermediate_act_fn(hidden_states)
        return hidden_states


class BertOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states, input_tensor):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


class BertLayer(nn.Module):
    def __init__(self, config, layer_num):
        super().__init__()
        self.config = config
        self.chunk_size_feed_forward = config.chunk_size_feed_forward
        self.seq_len_dim = 1
        self.attention = BertAttention(config)
        self.layer_num = layer_num
        if (
            self.config.add_cross_attention
            and layer_num % self.config.cross_attention_freq == 0
        ):
            self.crossattention = BertAttention(
                config, is_cross_attention=self.config.add_cross_attention
            )
            self.has_cross_attention = True
        else:
            self.has_cross_attention = False
        self.intermediate = BertIntermediate(config)
        self.output = BertOutput(config)

        self.intermediate_query = BertIntermediate(config)
        self.output_query = BertOutput(config)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_value=None,
        output_attentions=False,
        query_length=0,
    ):
        # decoder uni-directional self-attention cached key/values tuple is at positions 1,2
        self_attn_past_key_value = (
            past_key_value[:2] if past_key_value is not None else None
        )
        self_attention_outputs = self.attention(
            hidden_states,
            attention_mask,
            head_mask,
            output_attentions=output_attentions,
            past_key_value=self_attn_past_key_value,
        )
        attention_output = self_attention_outputs[0]
        outputs = self_attention_outputs[1:-1]

        present_key_value = self_attention_outputs[-1]

        if query_length > 0:
            query_attention_output = attention_output[:, :query_length, :]

            if self.has_cross_attention:
                assert (
                    encoder_hidden_states is not None
                ), "encoder_hidden_states must be given for cross-attention layers"
                cross_attention_outputs = self.crossattention(
                    query_attention_output,
                    attention_mask,
                    head_mask,
                    encoder_hidden_states,
                    encoder_attention_mask,
                    output_attentions=output_attentions,
                )
                query_attention_output = cross_attention_outputs[0]
                outputs = (
                    outputs + cross_attention_outputs[1:-1]
                )  # add cross attentions if we output attention weights

            layer_output = apply_chunking_to_forward(
                self.feed_forward_chunk_query,
                self.chunk_size_feed_forward,
                self.seq_len_dim,
                query_attention_output,
            )
            if attention_output.shape[1] > query_length:
                layer_output_text = apply_chunking_to_forward(
                    self.feed_forward_chunk,
                    self.chunk_size_feed_forward,
                    self.seq_len_dim,
                    attention_output[:, query_length:, :],
                )
                layer_output = torch.cat([layer_output, layer_output_text], dim=1)
        else:
            layer_output = apply_chunking_to_forward(
                self.feed_forward_chunk,
                self.chunk_size_feed_forward,
                self.seq_len_dim,
                attention_output,
            )
        outputs = (layer_output,) + outputs

        outputs = outputs + (present_key_value,)

        return outputs

    def feed_forward_chunk(self, attention_output):
        intermediate_output = self.intermediate(attention_output)
        layer_output = self.output(intermediate_output, attention_output)
        return layer_output

    def feed_forward_chunk_query(self, attention_output):
        intermediate_output = self.intermediate_query(attention_output)
        layer_output = self.output_query(intermediate_output, attention_output)
        return layer_output


class BertEncoder(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.layer = nn.ModuleList(
            [BertLayer(config, i) for i in range(config.num_hidden_layers)]
        )

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_values=None,
        use_cache=None,
        output_attentions=False,
        output_hidden_states=False,
        return_dict=True,
        query_length=0,
    ):
        all_hidden_states = () if output_hidden_states else None
        all_self_attentions = () if output_attentions else None
        all_cross_attentions = (
            () if output_attentions and self.config.add_cross_attention else None
        )

        next_decoder_cache = () if use_cache else None

        for i in range(self.config.num_hidden_layers):
            layer_module = self.layer[i]
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            layer_head_mask = head_mask[i] if head_mask is not None else None
            past_key_value = past_key_values[i] if past_key_values is not None else None

            if getattr(self.config, "gradient_checkpointing", False) and self.training:

                if use_cache:
                    logger.warn(
                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                    )
                    use_cache = False

                def create_custom_forward(module):
                    def custom_forward(*inputs):
                        return module(
                            *inputs, past_key_value, output_attentions, query_length
                        )

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(layer_module),
                    hidden_states,
                    attention_mask,
                    layer_head_mask,
                    encoder_hidden_states,
                    encoder_attention_mask,
                )
            else:
                layer_outputs = layer_module(
                    hidden_states,
                    attention_mask,
                    layer_head_mask,
                    encoder_hidden_states,
                    encoder_attention_mask,
                    past_key_value,
                    output_attentions,
                    query_length,
                )

            hidden_states = layer_outputs[0]
            if use_cache:
                next_decoder_cache += (layer_outputs[-1],)
            if output_attentions:
                all_self_attentions = all_self_attentions + (layer_outputs[1],)
                all_cross_attentions = all_cross_attentions + (layer_outputs[2],)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        if not return_dict:
            return tuple(
                v
                for v in [
                    hidden_states,
                    next_decoder_cache,
                    all_hidden_states,
                    all_self_attentions,
                    all_cross_attentions,
                ]
                if v is not None
            )
        return BaseModelOutputWithPastAndCrossAttentions(
            last_hidden_state=hidden_states,
            past_key_values=next_decoder_cache,
            hidden_states=all_hidden_states,
            attentions=all_self_attentions,
            cross_attentions=all_cross_attentions,
        )


class BertPooler(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.activation = nn.Tanh()

    def forward(self, hidden_states):
        # We "pool" the model by simply taking the hidden state corresponding
        # to the first token.
        first_token_tensor = hidden_states[:, 0]
        pooled_output = self.dense(first_token_tensor)
        pooled_output = self.activation(pooled_output)
        return pooled_output


class BertPredictionHeadTransform(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        if isinstance(config.hidden_act, str):
            self.transform_act_fn = ACT2FN[config.hidden_act]
        else:
            self.transform_act_fn = config.hidden_act
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)

    def forward(self, hidden_states):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.transform_act_fn(hidden_states)
        hidden_states = self.LayerNorm(hidden_states)
        return hidden_states


class BertLMPredictionHead(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.transform = BertPredictionHeadTransform(config)

        # The output weights are the same as the input embeddings, but there is
        # an output-only bias for each token.
        self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

        self.bias = nn.Parameter(torch.zeros(config.vocab_size))

        # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
        self.decoder.bias = self.bias

    def forward(self, hidden_states):
        hidden_states = self.transform(hidden_states)
        hidden_states = self.decoder(hidden_states)
        return hidden_states


class BertOnlyMLMHead(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.predictions = BertLMPredictionHead(config)

    def forward(self, sequence_output):
        prediction_scores = self.predictions(sequence_output)
        return prediction_scores


class BertPreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = BertConfig
    base_model_prefix = "bert"
    _keys_to_ignore_on_load_missing = [r"position_ids"]

    def _init_weights(self, module):
        """Initialize the weights"""
        if isinstance(module, (nn.Linear, nn.Embedding)):
            # Slightly different from the TF version which uses truncated_normal for initialization
            # cf https://github.com/pytorch/pytorch/pull/5617
            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
        elif isinstance(module, nn.LayerNorm):
            module.bias.data.zero_()
            module.weight.data.fill_(1.0)
        if isinstance(module, nn.Linear) and module.bias is not None:
            module.bias.data.zero_()


class BertModel(BertPreTrainedModel):
    """
    The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
    cross-attention is added between the self-attention layers, following the architecture described in `Attention is
    all you need <https://arxiv.org/abs/1706.03762>`__ by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
    Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
    argument and :obj:`add_cross_attention` set to :obj:`True`; an :obj:`encoder_hidden_states` is then expected as an
    input to the forward pass.
    """

    def __init__(self, config, add_pooling_layer=False):
        super().__init__(config)
        self.config = config

        self.embeddings = BertEmbeddings(config)

        self.encoder = BertEncoder(config)

        self.pooler = BertPooler(config) if add_pooling_layer else None

        self.init_weights()

    def get_input_embeddings(self):
        return self.embeddings.word_embeddings

    def set_input_embeddings(self, value):
        self.embeddings.word_embeddings = value

    def _prune_heads(self, heads_to_prune):
        """
        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
        class PreTrainedModel
        """
        for layer, heads in heads_to_prune.items():
            self.encoder.layer[layer].attention.prune_heads(heads)

    def get_extended_attention_mask(
        self,
        attention_mask: Tensor,
        input_shape: Tuple[int],
        device: device,
        is_decoder: bool,
        has_query: bool = False,
    ) -> Tensor:
        """
        Makes broadcastable attention and causal masks so that future and masked tokens are ignored.

        Arguments:
            attention_mask (:obj:`torch.Tensor`):
                Mask with ones indicating tokens to attend to, zeros for tokens to ignore.
            input_shape (:obj:`Tuple[int]`):
                The shape of the input to the model.
            device: (:obj:`torch.device`):
                The device of the input to the model.

        Returns:
            :obj:`torch.Tensor` The extended attention mask, with a the same dtype as :obj:`attention_mask.dtype`.
        """
        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
        # ourselves in which case we just need to make it broadcastable to all heads.
        if attention_mask.dim() == 3:
            extended_attention_mask = attention_mask[:, None, :, :]
        elif attention_mask.dim() == 2:
            # Provided a padding mask of dimensions [batch_size, seq_length]
            # - if the model is a decoder, apply a causal mask in addition to the padding mask
            # - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
            if is_decoder:
                batch_size, seq_length = input_shape

                seq_ids = torch.arange(seq_length, device=device)
                causal_mask = (
                    seq_ids[None, None, :].repeat(batch_size, seq_length, 1)
                    <= seq_ids[None, :, None]
                )

                # add a prefix ones mask to the causal mask
                # causal and attention masks must have same type with pytorch version < 1.3
                causal_mask = causal_mask.to(attention_mask.dtype)

                if causal_mask.shape[1] < attention_mask.shape[1]:
                    prefix_seq_len = attention_mask.shape[1] - causal_mask.shape[1]
                    if has_query:  # UniLM style attention mask
                        causal_mask = torch.cat(
                            [
                                torch.zeros(
                                    (batch_size, prefix_seq_len, seq_length),
                                    device=device,
                                    dtype=causal_mask.dtype,
                                ),
                                causal_mask,
                            ],
                            axis=1,
                        )
                    causal_mask = torch.cat(
                        [
                            torch.ones(
                                (batch_size, causal_mask.shape[1], prefix_seq_len),
                                device=device,
                                dtype=causal_mask.dtype,
                            ),
                            causal_mask,
                        ],
                        axis=-1,
                    )
                extended_attention_mask = (
                    causal_mask[:, None, :, :] * attention_mask[:, None, None, :]
                )
            else:
                extended_attention_mask = attention_mask[:, None, None, :]
        else:
            raise ValueError(
                "Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
                    input_shape, attention_mask.shape
                )
            )

        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
        # masked positions, this operation will create a tensor which is 0.0 for
        # positions we want to attend and -10000.0 for masked positions.
        # Since we are adding it to the raw scores before the softmax, this is
        # effectively the same as removing these entirely.
        extended_attention_mask = extended_attention_mask.to(
            dtype=self.dtype
        )  # fp16 compatibility
        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
        return extended_attention_mask

    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        position_ids=None,
        head_mask=None,
        query_embeds=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_values=None,
        use_cache=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
        is_decoder=False,
    ):
        r"""
        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
            the model is configured as a decoder.
        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
            the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
            - 1 for tokens that are **not masked**,
            - 0 for tokens that are **masked**.
        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
        use_cache (:obj:`bool`, `optional`):
            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
            decoding (see :obj:`past_key_values`).
        """
        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
        )

        # use_cache = use_cache if use_cache is not None else self.config.use_cache

        if input_ids is None:
            assert (
                query_embeds is not None
            ), "You have to specify query_embeds when input_ids is None"

        # past_key_values_length
        past_key_values_length = (
            past_key_values[0][0].shape[2] - self.config.query_length
            if past_key_values is not None
            else 0
        )

        query_length = query_embeds.shape[1] if query_embeds is not None else 0

        embedding_output = self.embeddings(
            input_ids=input_ids,
            position_ids=position_ids,
            query_embeds=query_embeds,
            past_key_values_length=past_key_values_length,
        )

        input_shape = embedding_output.size()[:-1]
        batch_size, seq_length = input_shape
        device = embedding_output.device

        if attention_mask is None:
            attention_mask = torch.ones(
                ((batch_size, seq_length + past_key_values_length)), device=device
            )

        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
        # ourselves in which case we just need to make it broadcastable to all heads.
        if is_decoder:
            extended_attention_mask = self.get_extended_attention_mask(
                attention_mask,
                input_ids.shape,
                device,
                is_decoder,
                has_query=(query_embeds is not None),
            )
        else:
            extended_attention_mask = self.get_extended_attention_mask(
                attention_mask, input_shape, device, is_decoder
            )

        # If a 2D or 3D attention mask is provided for the cross-attention
        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
        if encoder_hidden_states is not None:
            if type(encoder_hidden_states) == list:
                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[
                    0
                ].size()
            else:
                (
                    encoder_batch_size,
                    encoder_sequence_length,
                    _,
                ) = encoder_hidden_states.size()
            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)

            if type(encoder_attention_mask) == list:
                encoder_extended_attention_mask = [
                    self.invert_attention_mask(mask) for mask in encoder_attention_mask
                ]
            elif encoder_attention_mask is None:
                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
                encoder_extended_attention_mask = self.invert_attention_mask(
                    encoder_attention_mask
                )
            else:
                encoder_extended_attention_mask = self.invert_attention_mask(
                    encoder_attention_mask
                )
        else:
            encoder_extended_attention_mask = None

        # Prepare head mask if needed
        # 1.0 in head_mask indicate we keep the head
        # attention_probs has shape bsz x n_heads x N x N
        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)

        encoder_outputs = self.encoder(
            embedding_output,
            attention_mask=extended_attention_mask,
            head_mask=head_mask,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_extended_attention_mask,
            past_key_values=past_key_values,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            query_length=query_length,
        )
        sequence_output = encoder_outputs[0]
        pooled_output = (
            self.pooler(sequence_output) if self.pooler is not None else None
        )

        if not return_dict:
            return (sequence_output, pooled_output) + encoder_outputs[1:]

        return BaseModelOutputWithPoolingAndCrossAttentions(
            last_hidden_state=sequence_output,
            pooler_output=pooled_output,
            past_key_values=encoder_outputs.past_key_values,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
            cross_attentions=encoder_outputs.cross_attentions,
        )


class BertLMHeadModel(BertPreTrainedModel):

    _keys_to_ignore_on_load_unexpected = [r"pooler"]
    _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias"]

    def __init__(self, config):
        super().__init__(config)

        self.bert = BertModel(config, add_pooling_layer=False)
        self.cls = BertOnlyMLMHead(config)

        self.init_weights()

    def get_output_embeddings(self):
        return self.cls.predictions.decoder

    def set_output_embeddings(self, new_embeddings):
        self.cls.predictions.decoder = new_embeddings

    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        position_ids=None,
        head_mask=None,
        query_embeds=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        labels=None,
        past_key_values=None,
        use_cache=True,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
        return_logits=False,
        is_decoder=True,
        reduction="mean",
    ):
        r"""
        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
            the model is configured as a decoder.
        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
            the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
            - 1 for tokens that are **not masked**,
            - 0 for tokens that are **masked**.
        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
            Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in
            ``[-100, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are
            ignored (masked), the loss is only computed for the tokens with labels n ``[0, ..., config.vocab_size]``
        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
        use_cache (:obj:`bool`, `optional`):
            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
            decoding (see :obj:`past_key_values`).
        Returns:
        Example::
            >>> from transformers import BertTokenizer, BertLMHeadModel, BertConfig
            >>> import torch
            >>> tokenizer = BertTokenizer.from_pretrained('bert-base-cased')
            >>> config = BertConfig.from_pretrained("bert-base-cased")
            >>> model = BertLMHeadModel.from_pretrained('bert-base-cased', config=config)
            >>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
            >>> outputs = model(**inputs)
            >>> prediction_logits = outputs.logits
        """
        return_dict = (
            return_dict if return_dict is not None else self.config.use_return_dict
        )
        if labels is not None:
            use_cache = False
        if past_key_values is not None:
            query_embeds = None

        outputs = self.bert(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            head_mask=head_mask,
            query_embeds=query_embeds,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_attention_mask,
            past_key_values=past_key_values,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            is_decoder=is_decoder,
        )

        sequence_output = outputs[0]
        if query_embeds is not None:
            sequence_output = outputs[0][:, query_embeds.shape[1] :, :]

        prediction_scores = self.cls(sequence_output)

        if return_logits:
            return prediction_scores[:, :-1, :].contiguous()

        lm_loss = None
        if labels is not None:
            # we are doing next-token prediction; shift prediction scores and input ids by one
            shifted_prediction_scores = prediction_scores[:, :-1, :].contiguous()
            labels = labels[:, 1:].contiguous()
            loss_fct = CrossEntropyLoss(reduction=reduction, label_smoothing=0.1)
            lm_loss = loss_fct(
                shifted_prediction_scores.view(-1, self.config.vocab_size),
                labels.view(-1),
            )
            if reduction == "none":
                lm_loss = lm_loss.view(prediction_scores.size(0), -1).sum(1)

        if not return_dict:
            output = (prediction_scores,) + outputs[2:]
            return ((lm_loss,) + output) if lm_loss is not None else output

        return CausalLMOutputWithCrossAttentions(
            loss=lm_loss,
            logits=prediction_scores,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
            cross_attentions=outputs.cross_attentions,
        )

    def prepare_inputs_for_generation(
        self, input_ids, query_embeds, past=None, attention_mask=None, **model_kwargs
    ):
        # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
        if attention_mask is None:
            attention_mask = input_ids.new_ones(input_ids.shape)
        query_mask = input_ids.new_ones(query_embeds.shape[:-1])
        attention_mask = torch.cat([query_mask, attention_mask], dim=-1)

        # cut decoder_input_ids if past is used
        if past is not None:
            input_ids = input_ids[:, -1:]

        return {
            "input_ids": input_ids,
            "query_embeds": query_embeds,
            "attention_mask": attention_mask,
            "past_key_values": past,
            "encoder_hidden_states": model_kwargs.get("encoder_hidden_states", None),
            "encoder_attention_mask": model_kwargs.get("encoder_attention_mask", None),
            "is_decoder": True,
        }

    def _reorder_cache(self, past, beam_idx):
        reordered_past = ()
        for layer_past in past:
            reordered_past += (
                tuple(
                    past_state.index_select(0, beam_idx) for past_state in layer_past
                ),
            )
        return reordered_past


class BertForMaskedLM(BertPreTrainedModel):

    _keys_to_ignore_on_load_unexpected = [r"pooler"]
    _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias"]

    def __init__(self, config):
        super().__init__(config)

        self.bert = BertModel(config, add_pooling_layer=False)
        self.cls = BertOnlyMLMHead(config)

        self.init_weights()

    def get_output_embeddings(self):
        return self.cls.predictions.decoder

    def set_output_embeddings(self, new_embeddings):
        self.cls.predictions.decoder = new_embeddings

    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        position_ids=None,
        head_mask=None,
        query_embeds=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        labels=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
        return_logits=False,
        is_decoder=False,
    ):
        r"""
        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
            Labels for computing the masked language modeling loss. Indices should be in ``[-100, 0, ...,
            config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are ignored
            (masked), the loss is only computed for the tokens with labels in ``[0, ..., config.vocab_size]``
        """

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

        outputs = self.bert(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            head_mask=head_mask,
            query_embeds=query_embeds,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_attention_mask,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            is_decoder=is_decoder,
        )

        if query_embeds is not None:
            sequence_output = outputs[0][:, query_embeds.shape[1] :, :]
        prediction_scores = self.cls(sequence_output)

        if return_logits:
            return prediction_scores

        masked_lm_loss = None
        if labels is not None:
            loss_fct = CrossEntropyLoss()  # -100 index = padding token
            masked_lm_loss = loss_fct(
                prediction_scores.view(-1, self.config.vocab_size), labels.view(-1)
            )

        if not return_dict:
            output = (prediction_scores,) + outputs[2:]
            return (
                ((masked_lm_loss,) + output) if masked_lm_loss is not None else output
            )

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

# ------------------------------------------------------ #

class BEATsConfig:
    def __init__(self, cfg=None):
        # update the default values to BEATs_iter3_plus_AS2M_finetuned_on_AS2M_cpt2.pt
        self.input_patch_size: int = 16  # path size of patch embedding
        self.embed_dim: int = 512  # patch embedding dimension
        self.conv_bias: bool = False  # include bias in conv encoder

        self.encoder_layers: int = 12  # num encoder layers in the transformer
        self.encoder_embed_dim: int = 768  # encoder embedding dimension
        self.encoder_ffn_embed_dim: int = 3072  # encoder embedding dimension for FFN
        self.encoder_attention_heads: int = 12  # num encoder attention heads
        self.activation_fn: str = "gelu"  # activation function to use

        self.layer_wise_gradient_decay_ratio: float = 0.6  # ratio for layer-wise gradient decay
        self.layer_norm_first: bool = False  # apply layernorm first in the transformer
        self.deep_norm: bool = True  # apply deep_norm first in the transformer

        # dropouts
        self.dropout: float = 0.0  # dropout probability for the transformer
        self.attention_dropout: float = 0.0  # dropout probability for attention weights
        self.activation_dropout: float = 0.0  # dropout probability after activation in FFN
        self.encoder_layerdrop: float = 0.05  # probability of dropping a tarnsformer layer
        self.dropout_input: float = 0.0  # dropout to apply to the input (after feat extr)

        # positional embeddings
        self.conv_pos: int = 128  # number of filters for convolutional positional embeddings
        self.conv_pos_groups: int = 16  # number of groups for convolutional positional embedding

        # relative position embedding
        self.relative_position_embedding: bool = True  # apply relative position embedding
        self.num_buckets: int = 320  # number of buckets for relative position embedding
        self.max_distance: int = 800  # maximum distance for relative position embedding
        self.gru_rel_pos: bool = True  # apply gated relative position embedding

        # label predictor
        self.finetuned_model: bool = True  # whether the model is a fine-tuned model.
        self.predictor_dropout: float = 0.0  # dropout probability for the predictor
        self.predictor_class: int = 527  # target class number for the predictor

        if cfg is not None:
            self.update(cfg)

    def update(self, cfg: dict):
        self.__dict__.update(cfg)


class BEATs(nn.Module):
    def __init__(
            self,
            cfg: BEATsConfig,
    ) -> None:
        super().__init__()
        logger.info(f"BEATs Config: {cfg.__dict__}")

        self.cfg = cfg

        self.embed = cfg.embed_dim
        self.post_extract_proj = (
            nn.Linear(self.embed, cfg.encoder_embed_dim)
            if self.embed != cfg.encoder_embed_dim
            else None
        )

        self.input_patch_size = cfg.input_patch_size
        self.patch_embedding = nn.Conv2d(1, self.embed, kernel_size=self.input_patch_size, stride=self.input_patch_size,
                                         bias=cfg.conv_bias)

        self.dropout_input = nn.Dropout(cfg.dropout_input)

        assert not cfg.deep_norm or not cfg.layer_norm_first
        self.encoder = TransformerEncoder(cfg)
        self.layer_norm = LayerNorm(self.embed)

        if cfg.finetuned_model:
            self.predictor_dropout = nn.Dropout(cfg.predictor_dropout)
            self.predictor = nn.Linear(cfg.encoder_embed_dim, cfg.predictor_class)
        else:
            self.predictor = None

    def forward_padding_mask(
            self,
            features: torch.Tensor,
            padding_mask: torch.Tensor,
    ) -> torch.Tensor:
        extra = padding_mask.size(1) % features.size(1)
        if extra > 0:
            padding_mask = padding_mask[:, :-extra]
        padding_mask = padding_mask.view(
            padding_mask.size(0), features.size(1), -1
        )
        padding_mask = padding_mask.all(-1)
        return padding_mask

    def preprocess(
            self,
            source: torch.Tensor,
            fbank_mean: float = 15.41663,
            fbank_std: float = 6.55582,
    ) -> torch.Tensor:
        fbanks = []
        for waveform in source:
            waveform = waveform.unsqueeze(0) * 2 ** 15
            fbank = ta_kaldi.fbank(waveform, num_mel_bins=128, sample_frequency=16000, frame_length=25, frame_shift=10)
            fbanks.append(fbank)
        fbank = torch.stack(fbanks, dim=0)
        fbank = (fbank - fbank_mean) / (2 * fbank_std)
        return fbank

    def extract_features(
            self,
            source: torch.Tensor,
            padding_mask: Optional[torch.Tensor] = None,
            fbank_mean: float = 15.41663,
            fbank_std: float = 6.55582,
            feature_only=False,
            torch_dtype=torch.float32
    ):
        fbank = self.preprocess(source, fbank_mean=fbank_mean, fbank_std=fbank_std).to(torch_dtype)

        if padding_mask is not None:
            padding_mask = self.forward_padding_mask(fbank, padding_mask)

        fbank = fbank.unsqueeze(1)
        features = self.patch_embedding(fbank)
        features = features.reshape(features.shape[0], features.shape[1], -1)
        features = features.transpose(1, 2)
        features = self.layer_norm(features)

        if padding_mask is not None:
            padding_mask = self.forward_padding_mask(features, padding_mask)

        if self.post_extract_proj is not None:
            features = self.post_extract_proj(features)

        x = self.dropout_input(features)

        x, layer_results = self.encoder(
            x,
            padding_mask=padding_mask,
        )

        if not feature_only and self.predictor is not None:
            x = self.predictor_dropout(x)
            logits = self.predictor(x)

            if padding_mask is not None and padding_mask.any():
                logits[padding_mask] = 0
                logits = logits.sum(dim=1)
                logits = logits / (~padding_mask).sum(dim=1).unsqueeze(-1).expand_as(logits)
            else:
                logits = logits.mean(dim=1)

            lprobs = torch.sigmoid(logits)

            return lprobs, padding_mask
        else:
            return x, padding_mask

class TransformerEncoder(nn.Module):
    def __init__(self, args):
        super().__init__()

        self.dropout = args.dropout
        self.embedding_dim = args.encoder_embed_dim

        self.pos_conv = nn.Conv1d(
            self.embedding_dim,
            self.embedding_dim,
            kernel_size=args.conv_pos,
            padding=args.conv_pos // 2,
            groups=args.conv_pos_groups,
        )
        dropout = 0
        std = math.sqrt((4 * (1.0 - dropout)) / (args.conv_pos * self.embedding_dim))
        nn.init.normal_(self.pos_conv.weight, mean=0, std=std)
        nn.init.constant_(self.pos_conv.bias, 0)

        self.pos_conv = nn.utils.weight_norm(self.pos_conv, name="weight", dim=2)
        self.pos_conv = nn.Sequential(self.pos_conv, SamePad(args.conv_pos), nn.GELU())

        if hasattr(args, "relative_position_embedding"):
            self.relative_position_embedding = args.relative_position_embedding
            self.num_buckets = args.num_buckets
            self.max_distance = args.max_distance
        else:
            self.relative_position_embedding = False
            self.num_buckets = 0
            self.max_distance = 0

        self.layers = nn.ModuleList(
            [
                TransformerSentenceEncoderLayer(
                    embedding_dim=self.embedding_dim,
                    ffn_embedding_dim=args.encoder_ffn_embed_dim,
                    num_attention_heads=args.encoder_attention_heads,
                    dropout=self.dropout,
                    attention_dropout=args.attention_dropout,
                    activation_dropout=args.activation_dropout,
                    activation_fn=args.activation_fn,
                    layer_norm_first=args.layer_norm_first,
                    deep_norm=args.deep_norm,
                    has_relative_attention_bias=self.relative_position_embedding,
                    num_buckets=self.num_buckets,
                    max_distance=self.max_distance,
                    gru_rel_pos=args.gru_rel_pos,
                    encoder_layers=args.encoder_layers,
                )
                for i in range(args.encoder_layers)
            ]
        )
        if self.relative_position_embedding:
            for i in range(1, args.encoder_layers):
                del self.layers[i].self_attn.relative_attention_bias
                self.layers[i].self_attn.relative_attention_bias = self.layers[0].self_attn.relative_attention_bias

        self.layer_norm_first = args.layer_norm_first
        self.layer_norm = LayerNorm(self.embedding_dim)
        self.layerdrop = args.encoder_layerdrop

        self.apply(init_bert_params)

        if args.deep_norm:
            deep_norm_beta = math.pow(8 * args.encoder_layers, -1 / 4)
            for i in range(args.encoder_layers):
                nn.init.xavier_normal_(self.layers[i].self_attn.k_proj.weight, gain=1)
                nn.init.xavier_normal_(self.layers[i].self_attn.v_proj.weight, gain=deep_norm_beta)
                nn.init.xavier_normal_(self.layers[i].self_attn.q_proj.weight, gain=1)
                nn.init.xavier_normal_(self.layers[i].self_attn.out_proj.weight, gain=deep_norm_beta)
                nn.init.xavier_normal_(self.layers[i].fc1.weight, gain=deep_norm_beta)
                nn.init.xavier_normal_(self.layers[i].fc2.weight, gain=deep_norm_beta)

        self.layer_wise_gradient_decay_ratio = getattr(args, "layer_wise_gradient_decay_ratio", 1)

    def forward(self, x, padding_mask=None, layer=None):
        x, layer_results = self.extract_features(x, padding_mask, layer)

        if self.layer_norm_first and layer is None:
            x = self.layer_norm(x)

        return x, layer_results

    def extract_features(self, x, padding_mask=None, tgt_layer=None):

        if padding_mask is not None:
            x[padding_mask] = 0

        x_conv = self.pos_conv(x.transpose(1, 2))
        x_conv = x_conv.transpose(1, 2)
        x = x + x_conv

        if not self.layer_norm_first:
            x = self.layer_norm(x)

        x = F.dropout(x, p=self.dropout, training=self.training)

        # B x T x C -> T x B x C
        x = x.transpose(0, 1)

        layer_results = []
        z = None
        if tgt_layer is not None:
            layer_results.append((x, z))
        r = None
        pos_bias = None
        for i, layer in enumerate(self.layers):
            if self.layer_wise_gradient_decay_ratio != 1.0:
                x = GradMultiply.apply(x, self.layer_wise_gradient_decay_ratio)
            dropout_probability = np.random.random()
            if not self.training or (dropout_probability > self.layerdrop):
                x, z, pos_bias = layer(x, self_attn_padding_mask=padding_mask, need_weights=False, pos_bias=pos_bias)
            if tgt_layer is not None:
                layer_results.append((x, z))
            if i == tgt_layer:
                r = x
                break

        if r is not None:
            x = r

        # T x B x C -> B x T x C
        x = x.transpose(0, 1)

        return x, layer_results

class TransformerSentenceEncoderLayer(nn.Module):
    def __init__(
            self,
            embedding_dim: float = 768,
            ffn_embedding_dim: float = 3072,
            num_attention_heads: float = 8,
            dropout: float = 0.1,
            attention_dropout: float = 0.1,
            activation_dropout: float = 0.1,
            activation_fn: str = "relu",
            layer_norm_first: bool = False,
            deep_norm: bool = False,
            has_relative_attention_bias: bool = False,
            num_buckets: int = 0,
            max_distance: int = 0,
            rescale_init: bool = False,
            gru_rel_pos: bool = False,
            encoder_layers: int = 0,
    ) -> None:

        super().__init__()
        self.embedding_dim = embedding_dim
        self.dropout = dropout
        self.activation_dropout = activation_dropout

        self.activation_name = activation_fn
        self.activation_fn = get_activation_fn(activation_fn)
        self.self_attn = MultiheadAttention(
            self.embedding_dim,
            num_attention_heads,
            dropout=attention_dropout,
            self_attention=True,
            has_relative_attention_bias=has_relative_attention_bias,
            num_buckets=num_buckets,
            max_distance=max_distance,
            rescale_init=rescale_init,
            gru_rel_pos=gru_rel_pos,
        )

        self.dropout1 = nn.Dropout(dropout)
        self.dropout2 = nn.Dropout(self.activation_dropout)
        self.dropout3 = nn.Dropout(dropout)

        self.layer_norm_first = layer_norm_first

        self.self_attn_layer_norm = LayerNorm(self.embedding_dim)

        if self.activation_name == "glu":
            self.fc1 = GLU_Linear(self.embedding_dim, ffn_embedding_dim, "swish")
        else:
            self.fc1 = nn.Linear(self.embedding_dim, ffn_embedding_dim)
        self.fc2 = nn.Linear(ffn_embedding_dim, self.embedding_dim)

        self.final_layer_norm = LayerNorm(self.embedding_dim)

        self.deep_norm = deep_norm
        if self.deep_norm:
            self.deep_norm_alpha = math.pow(2 * encoder_layers, 1 / 4)
        else:
            self.deep_norm_alpha = 1

    def forward(
            self,
            x: torch.Tensor,
            self_attn_mask: torch.Tensor = None,
            self_attn_padding_mask: torch.Tensor = None,
            need_weights: bool = False,
            pos_bias=None
    ):
        residual = x

        if self.layer_norm_first:
            x = self.self_attn_layer_norm(x)
            x, attn, pos_bias = self.self_attn(
                query=x,
                key=x,
                value=x,
                key_padding_mask=self_attn_padding_mask,
                need_weights=False,
                attn_mask=self_attn_mask,
                position_bias=pos_bias
            )
            x = self.dropout1(x)
            x = residual + x

            residual = x
            x = self.final_layer_norm(x)
            if self.activation_name == "glu":
                x = self.fc1(x)
            else:
                x = self.activation_fn(self.fc1(x))
            x = self.dropout2(x)
            x = self.fc2(x)
            x = self.dropout3(x)
            x = residual + x
        else:
            x, attn, pos_bias = self.self_attn(
                query=x,
                key=x,
                value=x,
                key_padding_mask=self_attn_padding_mask,
                need_weights=need_weights,
                attn_mask=self_attn_mask,
                position_bias=pos_bias
            )

            x = self.dropout1(x)
            x = residual * self.deep_norm_alpha + x

            x = self.self_attn_layer_norm(x)

            residual = x
            if self.activation_name == "glu":
                x = self.fc1(x)
            else:
                x = self.activation_fn(self.fc1(x))
            x = self.dropout2(x)
            x = self.fc2(x)
            x = self.dropout3(x)
            x = residual * self.deep_norm_alpha + x
            x = self.final_layer_norm(x)

        return x, attn, pos_bias


class MultiheadAttention(nn.Module):
    """Multi-headed attention.

    See "Attention Is All You Need" for more details.
    """

    def __init__(
            self,
            embed_dim,
            num_heads,
            kdim=None,
            vdim=None,
            dropout=0.0,
            bias=True,
            add_bias_kv=False,
            add_zero_attn=False,
            self_attention=False,
            encoder_decoder_attention=False,
            q_noise=0.0,
            qn_block_size=8,
            has_relative_attention_bias=False,
            num_buckets=32,
            max_distance=128,
            gru_rel_pos=False,
            rescale_init=False,
    ):
        super().__init__()
        self.embed_dim = embed_dim
        self.kdim = kdim if kdim is not None else embed_dim
        self.vdim = vdim if vdim is not None else embed_dim
        self.qkv_same_dim = self.kdim == embed_dim and self.vdim == embed_dim

        self.num_heads = num_heads
        self.dropout_module = nn.Dropout(dropout)

        self.has_relative_attention_bias = has_relative_attention_bias
        self.num_buckets = num_buckets
        self.max_distance = max_distance
        if self.has_relative_attention_bias:
            self.relative_attention_bias = nn.Embedding(num_buckets, num_heads)

        self.head_dim = embed_dim // num_heads
        self.q_head_dim = self.head_dim
        self.k_head_dim = self.head_dim
        assert (
                self.head_dim * num_heads == self.embed_dim
        ), "embed_dim must be divisible by num_heads"
        self.scaling = self.head_dim ** -0.5

        self.self_attention = self_attention
        self.encoder_decoder_attention = encoder_decoder_attention

        assert not self.self_attention or self.qkv_same_dim, (
            "Self-attention requires query, key and " "value to be of the same size"
        )

        k_bias = True
        if rescale_init:
            k_bias = False

        k_embed_dim = embed_dim
        q_embed_dim = embed_dim

        self.k_proj = quant_noise(
            nn.Linear(self.kdim, k_embed_dim, bias=k_bias), q_noise, qn_block_size
        )
        self.v_proj = quant_noise(
            nn.Linear(self.vdim, embed_dim, bias=bias), q_noise, qn_block_size
        )
        self.q_proj = quant_noise(
            nn.Linear(embed_dim, q_embed_dim, bias=bias), q_noise, qn_block_size
        )

        self.out_proj = quant_noise(
            nn.Linear(embed_dim, embed_dim, bias=bias), q_noise, qn_block_size
        )

        if add_bias_kv:
            self.bias_k = Parameter(torch.Tensor(1, 1, embed_dim))
            self.bias_v = Parameter(torch.Tensor(1, 1, embed_dim))
        else:
            self.bias_k = self.bias_v = None

        self.add_zero_attn = add_zero_attn

        self.gru_rel_pos = gru_rel_pos
        if self.gru_rel_pos:
            self.grep_linear = nn.Linear(self.q_head_dim, 8)
            self.grep_a = nn.Parameter(torch.ones(1, num_heads, 1, 1))

        self.reset_parameters()

    def reset_parameters(self):
        if self.qkv_same_dim:
            # Empirically observed the convergence to be much better with
            # the scaled initialization
            nn.init.xavier_uniform_(self.k_proj.weight, gain=1 / math.sqrt(2))
            nn.init.xavier_uniform_(self.v_proj.weight, gain=1 / math.sqrt(2))
            nn.init.xavier_uniform_(self.q_proj.weight, gain=1 / math.sqrt(2))
        else:
            nn.init.xavier_uniform_(self.k_proj.weight)
            nn.init.xavier_uniform_(self.v_proj.weight)
            nn.init.xavier_uniform_(self.q_proj.weight)

        nn.init.xavier_uniform_(self.out_proj.weight)
        if self.out_proj.bias is not None:
            nn.init.constant_(self.out_proj.bias, 0.0)
        if self.bias_k is not None:
            nn.init.xavier_normal_(self.bias_k)
        if self.bias_v is not None:
            nn.init.xavier_normal_(self.bias_v)
        if self.has_relative_attention_bias:
            nn.init.xavier_normal_(self.relative_attention_bias.weight)

    def _relative_positions_bucket(self, relative_positions, bidirectional=True):
        num_buckets = self.num_buckets
        max_distance = self.max_distance
        relative_buckets = 0

        if bidirectional:
            num_buckets = num_buckets // 2
            relative_buckets += (relative_positions > 0).to(torch.long) * num_buckets
            relative_positions = torch.abs(relative_positions)
        else:
            relative_positions = -torch.min(relative_positions, torch.zeros_like(relative_positions))

        max_exact = num_buckets // 2
        is_small = relative_positions < max_exact

        relative_postion_if_large = max_exact + (
                torch.log(relative_positions.float() / max_exact)
                / math.log(max_distance / max_exact)
                * (num_buckets - max_exact)
        ).to(torch.long)
        relative_postion_if_large = torch.min(
            relative_postion_if_large, torch.full_like(relative_postion_if_large, num_buckets - 1)
        )

        relative_buckets += torch.where(is_small, relative_positions, relative_postion_if_large)
        return relative_buckets

    def compute_bias(self, query_length, key_length):
        context_position = torch.arange(query_length, dtype=torch.long)[:, None]
        memory_position = torch.arange(key_length, dtype=torch.long)[None, :]
        relative_position = memory_position - context_position
        relative_position_bucket = self._relative_positions_bucket(
            relative_position,
            bidirectional=True
        )
        relative_position_bucket = relative_position_bucket.to(self.relative_attention_bias.weight.device)
        values = self.relative_attention_bias(relative_position_bucket)
        values = values.permute([2, 0, 1])
        return values

    def forward(
            self,
            query,
            key: Optional[Tensor],
            value: Optional[Tensor],
            key_padding_mask: Optional[Tensor] = None,
            incremental_state: Optional[Dict[str, Dict[str, Optional[Tensor]]]] = None,
            need_weights: bool = True,
            static_kv: bool = False,
            attn_mask: Optional[Tensor] = None,
            before_softmax: bool = False,
            need_head_weights: bool = False,
            position_bias: Optional[Tensor] = None
    ) -> Tuple[Tensor, Optional[Tensor], Optional[Tensor]]:
        """Input shape: Time x Batch x Channel

        Args:
            key_padding_mask (ByteTensor, optional): mask to exclude
                keys that are pads, of shape `(batch, src_len)`, where
                padding elements are indicated by 1s.
            need_weights (bool, optional): return the attention weights,
                averaged over heads (default: False).
            attn_mask (ByteTensor, optional): typically used to
                implement causal attention, where the mask prevents the
                attention from looking forward in time (default: None).
            before_softmax (bool, optional): return the raw attention
                weights and values before the attention softmax.
            need_head_weights (bool, optional): return the attention
                weights for each head. Implies *need_weights*. Default:
                return the average attention weights over all heads.
        """
        if need_head_weights:
            need_weights = True

        is_tpu = query.device.type == "xla"

        tgt_len, bsz, embed_dim = query.size()
        src_len = tgt_len
        assert embed_dim == self.embed_dim
        assert list(query.size()) == [tgt_len, bsz, embed_dim]
        if key is not None:
            src_len, key_bsz, _ = key.size()
            if not torch.jit.is_scripting():
                assert key_bsz == bsz
                assert value is not None
                assert src_len, bsz == value.shape[:2]

        if self.has_relative_attention_bias and position_bias is None:
            position_bias = self.compute_bias(tgt_len, src_len)
            position_bias = position_bias.unsqueeze(0).repeat(bsz, 1, 1, 1).view(bsz * self.num_heads, tgt_len, src_len)

        if incremental_state is not None:
            saved_state = self._get_input_buffer(incremental_state)
            if saved_state is not None and "prev_key" in saved_state:
                # previous time steps are cached - no need to recompute
                # key and value if they are static
                if static_kv:
                    assert self.encoder_decoder_attention and not self.self_attention
                    key = value = None
        else:
            saved_state = None

        if self.self_attention:
            q = self.q_proj(query)
            k = self.k_proj(query)
            v = self.v_proj(query)
        elif self.encoder_decoder_attention:
            # encoder-decoder attention
            q = self.q_proj(query)
            if key is None:
                assert value is None
                k = v = None
            else:
                k = self.k_proj(key)
                v = self.v_proj(key)

        else:
            assert key is not None and value is not None
            q = self.q_proj(query)
            k = self.k_proj(key)
            v = self.v_proj(value)
        q *= self.scaling
        alpha = 32
        q *= 1 / alpha

        if self.bias_k is not None:
            assert self.bias_v is not None
            k = torch.cat([k, self.bias_k.repeat(1, bsz, 1)])
            v = torch.cat([v, self.bias_v.repeat(1, bsz, 1)])
            if attn_mask is not None:
                attn_mask = torch.cat(
                    [attn_mask, attn_mask.new_zeros(attn_mask.size(0), 1)], dim=1
                )
            if key_padding_mask is not None:
                key_padding_mask = torch.cat(
                    [
                        key_padding_mask,
                        key_padding_mask.new_zeros(key_padding_mask.size(0), 1),
                    ],
                    dim=1,
                )

        q = (
            q.contiguous()
                .view(tgt_len, bsz * self.num_heads, self.q_head_dim)
                .transpose(0, 1)
        )
        if k is not None:
            k = (
                k.contiguous()
                    .view(-1, bsz * self.num_heads, self.k_head_dim)
                    .transpose(0, 1)
            )
        if v is not None:
            v = (
                v.contiguous()
                    .view(-1, bsz * self.num_heads, self.head_dim)
                    .transpose(0, 1)
            )

        if saved_state is not None:
            # saved states are stored with shape (bsz, num_heads, seq_len, head_dim)
            if "prev_key" in saved_state:
                _prev_key = saved_state["prev_key"]
                assert _prev_key is not None
                prev_key = _prev_key.view(bsz * self.num_heads, -1, self.head_dim)
                if static_kv:
                    k = prev_key
                else:
                    assert k is not None
                    k = torch.cat([prev_key, k], dim=1)
                src_len = k.size(1)
            if "prev_value" in saved_state:
                _prev_value = saved_state["prev_value"]
                assert _prev_value is not None
                prev_value = _prev_value.view(bsz * self.num_heads, -1, self.head_dim)
                if static_kv:
                    v = prev_value
                else:
                    assert v is not None
                    v = torch.cat([prev_value, v], dim=1)
            prev_key_padding_mask: Optional[Tensor] = None
            if "prev_key_padding_mask" in saved_state:
                prev_key_padding_mask = saved_state["prev_key_padding_mask"]
            assert k is not None and v is not None
            key_padding_mask = MultiheadAttention._append_prev_key_padding_mask(
                key_padding_mask=key_padding_mask,
                prev_key_padding_mask=prev_key_padding_mask,
                batch_size=bsz,
                src_len=k.size(1),
                static_kv=static_kv,
            )

            saved_state["prev_key"] = k.view(bsz, self.num_heads, -1, self.head_dim)
            saved_state["prev_value"] = v.view(bsz, self.num_heads, -1, self.head_dim)
            saved_state["prev_key_padding_mask"] = key_padding_mask
            # In this branch incremental_state is never None
            assert incremental_state is not None
            incremental_state = self._set_input_buffer(incremental_state, saved_state)
        assert k is not None
        assert k.size(1) == src_len

        # This is part of a workaround to get around fork/join parallelism
        # not supporting Optional types.
        if key_padding_mask is not None and key_padding_mask.dim() == 0:
            key_padding_mask = None

        if key_padding_mask is not None:
            assert key_padding_mask.size(0) == bsz
            assert key_padding_mask.size(1) == src_len

        if self.add_zero_attn:
            assert v is not None
            src_len += 1
            k = torch.cat([k, k.new_zeros((k.size(0), 1) + k.size()[2:])], dim=1)
            v = torch.cat([v, v.new_zeros((v.size(0), 1) + v.size()[2:])], dim=1)
            if attn_mask is not None:
                attn_mask = torch.cat(
                    [attn_mask, attn_mask.new_zeros(attn_mask.size(0), 1)], dim=1
                )
            if key_padding_mask is not None:
                key_padding_mask = torch.cat(
                    [
                        key_padding_mask,
                        torch.zeros(key_padding_mask.size(0), 1).type_as(
                            key_padding_mask
                        ),
                    ],
                    dim=1,
                )

        attn_weights = torch.bmm(q, k.transpose(1, 2))
        attn_weights = (attn_weights - attn_weights.max(dim=-1, keepdim=True)[0]) * alpha
        attn_weights = self.apply_sparse_mask(attn_weights, tgt_len, src_len, bsz)

        assert list(attn_weights.size()) == [bsz * self.num_heads, tgt_len, src_len]

        if attn_mask is not None:
            attn_mask = attn_mask.unsqueeze(0)
            attn_weights += attn_mask

        if key_padding_mask is not None:
            # don't attend to padding symbols
            attn_weights = attn_weights.view(bsz, self.num_heads, tgt_len, src_len)
            if not is_tpu:
                attn_weights = attn_weights.masked_fill(
                    key_padding_mask.unsqueeze(1).unsqueeze(2).to(torch.bool),
                    float("-inf"),
                )
            else:
                attn_weights = attn_weights.transpose(0, 2)
                attn_weights = attn_weights.masked_fill(key_padding_mask, float("-inf"))
                attn_weights = attn_weights.transpose(0, 2)
            attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)

        if before_softmax:
            return attn_weights, v, position_bias

        if position_bias is not None:
            attn_mask_rel_pos = position_bias
            if self.gru_rel_pos == 1:
                query_layer = q.view(bsz, self.num_heads, tgt_len, self.q_head_dim) * alpha / self.scaling
                _B, _H, _L, __ = query_layer.size()
                gate_a, gate_b = torch.sigmoid(self.grep_linear(query_layer).view(
                    _B, _H, _L, 2, 4).sum(-1, keepdim=False)).chunk(2, dim=-1)
                gate_a_1 = gate_a * (gate_b * self.grep_a - 1.0) + 2.0
                attn_mask_rel_pos = gate_a_1.view(bsz * self.num_heads, tgt_len, 1) * position_bias

            attn_mask_rel_pos = attn_mask_rel_pos.view(attn_weights.size())

            attn_weights = attn_weights + attn_mask_rel_pos

        attn_weights_float = F.softmax(
            attn_weights, dim=-1
        )
        attn_weights = attn_weights_float.type_as(attn_weights)
        attn_probs = self.dropout_module(attn_weights)

        assert v is not None
        attn = torch.bmm(attn_probs, v)
        assert list(attn.size()) == [bsz * self.num_heads, tgt_len, self.head_dim]
        attn = attn.transpose(0, 1).contiguous().view(tgt_len, bsz, embed_dim)
        attn = self.out_proj(attn)
        attn_weights: Optional[Tensor] = None
        if need_weights:
            attn_weights = attn_weights_float.view(
                bsz, self.num_heads, tgt_len, src_len
            ).transpose(1, 0)
            if not need_head_weights:
                # average attention weights over heads
                attn_weights = attn_weights.mean(dim=0)

        return attn, attn_weights, position_bias

    @staticmethod
    def _append_prev_key_padding_mask(
            key_padding_mask: Optional[Tensor],
            prev_key_padding_mask: Optional[Tensor],
            batch_size: int,
            src_len: int,
            static_kv: bool,
    ) -> Optional[Tensor]:
        # saved key padding masks have shape (bsz, seq_len)
        if prev_key_padding_mask is not None and static_kv:
            new_key_padding_mask = prev_key_padding_mask
        elif prev_key_padding_mask is not None and key_padding_mask is not None:
            new_key_padding_mask = torch.cat(
                [prev_key_padding_mask.float(), key_padding_mask.float()], dim=1
            )
        # During incremental decoding, as the padding token enters and
        # leaves the frame, there will be a time when prev or current
        # is None
        elif prev_key_padding_mask is not None:
            if src_len > prev_key_padding_mask.size(1):
                filler = torch.zeros(
                    (batch_size, src_len - prev_key_padding_mask.size(1)),
                    device=prev_key_padding_mask.device,
                )
                new_key_padding_mask = torch.cat(
                    [prev_key_padding_mask.float(), filler.float()], dim=1
                )
            else:
                new_key_padding_mask = prev_key_padding_mask.float()
        elif key_padding_mask is not None:
            if src_len > key_padding_mask.size(1):
                filler = torch.zeros(
                    (batch_size, src_len - key_padding_mask.size(1)),
                    device=key_padding_mask.device,
                )
                new_key_padding_mask = torch.cat(
                    [filler.float(), key_padding_mask.float()], dim=1
                )
            else:
                new_key_padding_mask = key_padding_mask.float()
        else:
            new_key_padding_mask = prev_key_padding_mask
        return new_key_padding_mask

    def _get_input_buffer(
            self, incremental_state: Optional[Dict[str, Dict[str, Optional[Tensor]]]]
    ) -> Dict[str, Optional[Tensor]]:
        result = self.get_incremental_state(incremental_state, "attn_state")
        if result is not None:
            return result
        else:
            empty_result: Dict[str, Optional[Tensor]] = {}
            return empty_result

    def _set_input_buffer(
            self,
            incremental_state: Dict[str, Dict[str, Optional[Tensor]]],
            buffer: Dict[str, Optional[Tensor]],
    ):
        return self.set_incremental_state(incremental_state, "attn_state", buffer)

    def apply_sparse_mask(self, attn_weights, tgt_len: int, src_len: int, bsz: int):
        return attn_weights


def init_bert_params(module):
    """
    Initialize the weights specific to the BERT Model.
    This overrides the default initializations depending on the specified arguments.
        1. If normal_init_linear_weights is set then weights of linear
           layer will be initialized using the normal distribution and
           bais will be set to the specified value.
        2. If normal_init_embed_weights is set then weights of embedding
           layer will be initialized using the normal distribution.
        3. If normal_init_proj_weights is set then weights of
           in_project_weight for MultiHeadAttention initialized using
           the normal distribution (to be validated).
    """

    def normal_(data):
        # with FSDP, module params will be on CUDA, so we cast them back to CPU
        # so that the RNG is consistent with and without FSDP
        data.copy_(
            data.cpu().normal_(mean=0.0, std=0.02).to(data.device)
        )

    if isinstance(module, nn.Linear):
        normal_(module.weight.data)
        if module.bias is not None:
            module.bias.data.zero_()
    if isinstance(module, nn.Embedding):
        normal_(module.weight.data)
        if module.padding_idx is not None:
            module.weight.data[module.padding_idx].zero_()
    if isinstance(module, MultiheadAttention):
        normal_(module.q_proj.weight.data)
        normal_(module.k_proj.weight.data)
        normal_(module.v_proj.weight.data)



class GradMultiply(torch.autograd.Function):
    @staticmethod
    def forward(ctx, x, scale):
        ctx.scale = scale
        res = x.new(x)
        return res

    @staticmethod
    def backward(ctx, grad):
        return grad * ctx.scale, None


class SamePad(nn.Module):
    def __init__(self, kernel_size, causal=False):
        super().__init__()
        if causal:
            self.remove = kernel_size - 1
        else:
            self.remove = 1 if kernel_size % 2 == 0 else 0

    def forward(self, x):
        if self.remove > 0:
            x = x[:, :, : -self.remove]
        return x


class Swish(nn.Module):
    def __init__(self):
        super(Swish, self).__init__()
        self.act = torch.nn.Sigmoid()

    def forward(self, x):
        return x * self.act(x)


class GLU_Linear(nn.Module):
    def __init__(self, input_dim, output_dim, glu_type="sigmoid", bias_in_glu=True):
        super(GLU_Linear, self).__init__()

        self.glu_type = glu_type
        self.output_dim = output_dim

        if glu_type == "sigmoid":
            self.glu_act = torch.nn.Sigmoid()
        elif glu_type == "swish":
            self.glu_act = Swish()
        elif glu_type == "relu":
            self.glu_act = torch.nn.ReLU()
        elif glu_type == "gelu":
            self.glu_act = torch.nn.GELU()

        if bias_in_glu:
            self.linear = nn.Linear(input_dim, output_dim * 2, True)
        else:
            self.linear = nn.Linear(input_dim, output_dim * 2, False)

    def forward(self, x):
        # to be consistent with GLU_Linear, we assume the input always has the #channel (#dim) in the last dimension of the tensor, so need to switch the dimension first for 1D-Conv case
        x = self.linear(x)

        if self.glu_type == "bilinear":
            x = (x[:, :, 0:self.output_dim] * x[:, :, self.output_dim:self.output_dim * 2])
        else:
            x = (x[:, :, 0:self.output_dim] * self.glu_act(x[:, :, self.output_dim:self.output_dim * 2]))

        return x


def gelu_accurate(x):
    if not hasattr(gelu_accurate, "_a"):
        gelu_accurate._a = math.sqrt(2 / math.pi)
    return (
        0.5 * x * (1 + torch.tanh(gelu_accurate._a * (x + 0.044715 * torch.pow(x, 3))))
    )


def gelu(x: torch.Tensor) -> torch.Tensor:
    return torch.nn.functional.gelu(x.float()).type_as(x)


def get_activation_fn(activation: str):
    """Returns the activation function corresponding to `activation`"""

    if activation == "relu":
        return F.relu
    elif activation == "gelu":
        return gelu
    elif activation == "gelu_fast":
        warnings.warn(
            "--activation-fn=gelu_fast has been renamed to gelu_accurate"
        )
        return gelu_accurate
    elif activation == "gelu_accurate":
        return gelu_accurate
    elif activation == "tanh":
        return torch.tanh
    elif activation == "linear":
        return lambda x: x
    elif activation == "glu":
        return lambda x: x
    else:
        raise RuntimeError("--activation-fn {} not supported".format(activation))


def quant_noise(module, p, block_size):
    """
    Wraps modules and applies quantization noise to the weights for
    subsequent quantization with Iterative Product Quantization as
    described in "Training with Quantization Noise for Extreme Model Compression"

    Args:
        - module: nn.Module
        - p: amount of Quantization Noise
        - block_size: size of the blocks for subsequent quantization with iPQ

    Remarks:
        - Module weights must have the right sizes wrt the block size
        - Only Linear, Embedding and Conv2d modules are supported for the moment
        - For more detail on how to quantize by blocks with convolutional weights,
          see "And the Bit Goes Down: Revisiting the Quantization of Neural Networks"
        - We implement the simplest form of noise here as stated in the paper
          which consists in randomly dropping blocks
    """

    # if no quantization noise, don't register hook
    if p <= 0:
        return module

    # supported modules
    assert isinstance(module, (nn.Linear, nn.Embedding, nn.Conv2d))

    # test whether module.weight has the right sizes wrt block_size
    is_conv = module.weight.ndim == 4

    # 2D matrix
    if not is_conv:
        assert (
            module.weight.size(1) % block_size == 0
        ), "Input features must be a multiple of block sizes"

    # 4D matrix
    else:
        # 1x1 convolutions
        if module.kernel_size == (1, 1):
            assert (
                module.in_channels % block_size == 0
            ), "Input channels must be a multiple of block sizes"
        # regular convolutions
        else:
            k = module.kernel_size[0] * module.kernel_size[1]
            assert k % block_size == 0, "Kernel size must be a multiple of block size"

    def _forward_pre_hook(mod, input):
        # no noise for evaluation
        if mod.training:
            if not is_conv:
                # gather weight and sizes
                weight = mod.weight
                in_features = weight.size(1)
                out_features = weight.size(0)

                # split weight matrix into blocks and randomly drop selected blocks
                mask = torch.zeros(
                    in_features // block_size * out_features, device=weight.device
                )
                mask.bernoulli_(p)
                mask = mask.repeat_interleave(block_size, -1).view(-1, in_features)

            else:
                # gather weight and sizes
                weight = mod.weight
                in_channels = mod.in_channels
                out_channels = mod.out_channels

                # split weight matrix into blocks and randomly drop selected blocks
                if mod.kernel_size == (1, 1):
                    mask = torch.zeros(
                        int(in_channels // block_size * out_channels),
                        device=weight.device,
                    )
                    mask.bernoulli_(p)
                    mask = mask.repeat_interleave(block_size, -1).view(-1, in_channels)
                else:
                    mask = torch.zeros(
                        weight.size(0), weight.size(1), device=weight.device
                    )
                    mask.bernoulli_(p)
                    mask = (
                        mask.unsqueeze(2)
                        .unsqueeze(3)
                        .repeat(1, 1, mod.kernel_size[0], mod.kernel_size[1])
                    )

            # scale weights and apply mask
            mask = mask.to(
                torch.bool
            )  # x.bool() is not currently supported in TorchScript
            s = 1 / (1 - p)
            mod.weight.data = s * weight.masked_fill(mask, 0)

    module.register_forward_pre_hook(_forward_pre_hook)
    return module


class TokenizersConfig:
    def __init__(self, cfg=None):
        self.input_patch_size: int = -1  # path size of patch embedding
        self.embed_dim: int = 512  # patch embedding dimension
        self.conv_bias: bool = False  # include bias in conv encoder

        self.encoder_layers: int = 12  # num encoder layers in the transformer
        self.encoder_embed_dim: int = 768  # encoder embedding dimension
        self.encoder_ffn_embed_dim: int = 3072  # encoder embedding dimension for FFN
        self.encoder_attention_heads: int = 12  # num encoder attention heads
        self.activation_fn: str = "gelu"  # activation function to use

        self.layer_norm_first: bool = False  # apply layernorm first in the transformer
        self.deep_norm: bool = False  # apply deep_norm first in the transformer

        # dropouts
        self.dropout: float = 0.1  # dropout probability for the transformer
        self.attention_dropout: float = 0.1  # dropout probability for attention weights
        self.activation_dropout: float = 0.0  # dropout probability after activation in FFN
        self.encoder_layerdrop: float = 0.0  # probability of dropping a tarnsformer layer
        self.dropout_input: float = 0.0  # dropout to apply to the input (after feat extr)

        # positional embeddings
        self.conv_pos: int = 128  # number of filters for convolutional positional embeddings
        self.conv_pos_groups: int = 16  # number of groups for convolutional positional embedding

        # relative position embedding
        self.relative_position_embedding: bool = False  # apply relative position embedding
        self.num_buckets: int = 320  # number of buckets for relative position embedding
        self.max_distance: int = 1280  # maximum distance for relative position embedding
        self.gru_rel_pos: bool = False  # apply gated relative position embedding

        # quantizer
        self.quant_n: int = 1024 # codebook number in quantizer
        self.quant_dim: int = 256    # codebook dimension in quantizer

        if cfg is not None:
            self.update(cfg)

    def update(self, cfg: dict):
        self.__dict__.update(cfg)


class Tokenizers(nn.Module):
    def __init__(
            self,
            cfg: TokenizersConfig,
    ) -> None:
        super().__init__()
        logger.info(f"Tokenizers Config: {cfg.__dict__}")

        self.cfg = cfg

        self.embed = cfg.embed_dim
        self.post_extract_proj = (
            nn.Linear(self.embed, cfg.encoder_embed_dim)
            if self.embed != cfg.encoder_embed_dim
            else None
        )

        self.input_patch_size = cfg.input_patch_size
        self.patch_embedding = nn.Conv2d(1, self.embed, kernel_size=self.input_patch_size, stride=self.input_patch_size,
                                         bias=cfg.conv_bias)

        self.dropout_input = nn.Dropout(cfg.dropout_input)

        assert not cfg.deep_norm or not cfg.layer_norm_first
        self.encoder = TransformerEncoder(cfg)
        self.layer_norm = LayerNorm(self.embed)

        self.quantize = NormEMAVectorQuantizer(
            n_embed=cfg.quant_n, embedding_dim=cfg.quant_dim, beta=1.0, kmeans_init=True, decay=0.99,
        )
        self.quant_n = cfg.quant_n
        self.quantize_layer = nn.Sequential(
            nn.Linear(cfg.encoder_embed_dim, cfg.encoder_embed_dim),
            nn.Tanh(),
            nn.Linear(cfg.encoder_embed_dim, cfg.quant_dim)  # for quantize
        )

    def forward_padding_mask(
            self,
            features: torch.Tensor,
            padding_mask: torch.Tensor,
    ) -> torch.Tensor:
        extra = padding_mask.size(1) % features.size(1)
        if extra > 0:
            padding_mask = padding_mask[:, :-extra]
        padding_mask = padding_mask.view(
            padding_mask.size(0), features.size(1), -1
        )
        padding_mask = padding_mask.all(-1)
        return padding_mask

    def preprocess(
            self,
            source: torch.Tensor,
            fbank_mean: float = 15.41663,
            fbank_std: float = 6.55582,
    ) -> torch.Tensor:
        fbanks = []
        for waveform in source:
            waveform = waveform.unsqueeze(0) * 2 ** 15
            fbank = ta_kaldi.fbank(waveform, num_mel_bins=128, sample_frequency=16000, frame_length=25, frame_shift=10)
            fbanks.append(fbank)
        fbank = torch.stack(fbanks, dim=0)
        fbank = (fbank - fbank_mean) / (2 * fbank_std)
        return fbank

    def extract_labels(
            self,
            source: torch.Tensor,
            padding_mask: Optional[torch.Tensor] = None,
            fbank_mean: float = 15.41663,
            fbank_std: float = 6.55582,
    ):
        fbank = self.preprocess(source, fbank_mean=fbank_mean, fbank_std=fbank_std)

        if padding_mask is not None:
            padding_mask = self.forward_padding_mask(fbank, padding_mask)

        fbank = fbank.unsqueeze(1)
        features = self.patch_embedding(fbank)
        features = features.reshape(features.shape[0], features.shape[1], -1)
        features = features.transpose(1, 2)
        features = self.layer_norm(features)

        if padding_mask is not None:
            padding_mask = self.forward_padding_mask(features, padding_mask)

        if self.post_extract_proj is not None:
            features = self.post_extract_proj(features)

        x = self.dropout_input(features)

        x, layer_results = self.encoder(
            x,
            padding_mask=padding_mask,
        )

        quantize_input = self.quantize_layer(x)
        quantize_feature, embed_loss, embed_ind = self.quantize(quantize_input)

        return embed_ind


def l2norm(t):
    return F.normalize(t, p=2, dim=-1)


def ema_inplace(moving_avg, new, decay):
    moving_avg.data.mul_(decay).add_(new, alpha=(1 - decay))


def sample_vectors(samples, num):
    num_samples, device = samples.shape[0], samples.device

    if num_samples >= num:
        indices = torch.randperm(num_samples, device=device)[:num]
    else:
        indices = torch.randint(0, num_samples, (num,), device=device)

    return samples[indices]


def kmeans(samples, num_clusters, num_iters=10, use_cosine_sim=False):
    dim, dtype, device = samples.shape[-1], samples.dtype, samples.device

    means = sample_vectors(samples, num_clusters)

    for _ in range(num_iters):
        if use_cosine_sim:
            dists = samples @ means.t()
        else:
            diffs = rearrange(samples, 'n d -> n () d') \
                    - rearrange(means, 'c d -> () c d')
            dists = -(diffs ** 2).sum(dim=-1)

        buckets = dists.max(dim=-1).indices
        bins = torch.bincount(buckets, minlength=num_clusters)
        zero_mask = bins == 0
        bins_min_clamped = bins.masked_fill(zero_mask, 1)

        new_means = buckets.new_zeros(num_clusters, dim, dtype=dtype)
        new_means.scatter_add_(0, repeat(buckets, 'n -> n d', d=dim), samples)
        new_means = new_means / bins_min_clamped[..., None]

        if use_cosine_sim:
            new_means = l2norm(new_means)

        means = torch.where(zero_mask[..., None], means, new_means)

    return means, bins


class EmbeddingEMA(nn.Module):
    def __init__(self, num_tokens, codebook_dim, decay=0.99, eps=1e-5, kmeans_init=True, codebook_init_path=''):
        super().__init__()
        self.num_tokens = num_tokens
        self.codebook_dim = codebook_dim
        self.decay = decay
        self.eps = eps
        if codebook_init_path == '':
            if not kmeans_init:
                weight = torch.randn(num_tokens, codebook_dim)
                weight = l2norm(weight)
            else:
                weight = torch.zeros(num_tokens, codebook_dim)
            self.register_buffer('initted', torch.Tensor([not kmeans_init]))
        else:
            print(f"load init codebook weight from {codebook_init_path}")
            codebook_ckpt_weight = torch.load(codebook_init_path, map_location='cpu')
            weight = codebook_ckpt_weight.clone()
            self.register_buffer('initted', torch.Tensor([True]))

        self.weight = nn.Parameter(weight, requires_grad=False)
        self.cluster_size = nn.Parameter(torch.zeros(num_tokens), requires_grad=False)
        self.embed_avg = nn.Parameter(weight.clone(), requires_grad=False)
        # self.register_buffer('initted', torch.Tensor([not kmeans_init]))
        self.update = True

    @torch.jit.ignore
    def init_embed_(self, data):
        if self.initted:
            return
        print("Performing Kemans init for codebook")
        embed, cluster_size = kmeans(data, self.num_tokens, 10, use_cosine_sim=True)
        self.weight.data.copy_(embed)
        self.cluster_size.data.copy_(cluster_size)
        self.initted.data.copy_(torch.Tensor([True]))

    def forward(self, embed_id):
        return F.embedding(embed_id, self.weight)

    def cluster_size_ema_update(self, new_cluster_size):
        self.cluster_size.data.mul_(self.decay).add_(new_cluster_size, alpha=1 - self.decay)

    def embed_avg_ema_update(self, new_embed_avg):
        self.embed_avg.data.mul_(self.decay).add_(new_embed_avg, alpha=1 - self.decay)

    def weight_update(self, num_tokens):
        n = self.cluster_size.sum()
        smoothed_cluster_size = (
                (self.cluster_size + self.eps) / (n + num_tokens * self.eps) * n
        )
        # normalize embedding average with smoothed cluster size
        embed_normalized = self.embed_avg / smoothed_cluster_size.unsqueeze(1)
        # embed_normalized = l2norm(self.embed_avg / smoothed_cluster_size.unsqueeze(1))
        self.weight.data.copy_(embed_normalized)


def norm_ema_inplace(moving_avg, new, decay):
    moving_avg.data.mul_(decay).add_(new, alpha=(1 - decay))
    moving_avg.data.copy_(l2norm(moving_avg.data))


class NormEMAVectorQuantizer(nn.Module):
    def __init__(self, n_embed, embedding_dim, beta, decay=0.99, eps=1e-5,
                 statistic_code_usage=True, kmeans_init=False, codebook_init_path=''):
        super().__init__()
        self.codebook_dim = embedding_dim
        self.num_tokens = n_embed
        self.beta = beta
        self.decay = decay

        # learnable = True if orthogonal_reg_weight > 0 else False
        self.embedding = EmbeddingEMA(self.num_tokens, self.codebook_dim, decay, eps, kmeans_init, codebook_init_path)

        self.statistic_code_usage = statistic_code_usage
        if statistic_code_usage:
            self.register_buffer('cluster_size', torch.zeros(n_embed))
        if distributed.is_available() and distributed.is_initialized():
            print("ddp is enable, so use ddp_reduce to sync the statistic_code_usage for each gpu!")
            self.all_reduce_fn = distributed.all_reduce
        else:
            self.all_reduce_fn = nn.Identity()

    def reset_cluster_size(self, device):
        if self.statistic_code_usage:
            self.register_buffer('cluster_size', torch.zeros(self.num_tokens))
            self.cluster_size = self.cluster_size.to(device)

    def forward(self, z):
        # reshape z -> (batch, height, width, channel) and flatten
        # z, 'b c h w -> b h w c'
        # z = rearrange(z, 'b c h w -> b h w c')
        # z = z.transpose(1, 2)
        z = l2norm(z)
        z_flattened = z.reshape(-1, self.codebook_dim)

        self.embedding.init_embed_(z_flattened)

        d = z_flattened.pow(2).sum(dim=1, keepdim=True) + \
            self.embedding.weight.pow(2).sum(dim=1) - 2 * \
            torch.einsum('bd,nd->bn', z_flattened, self.embedding.weight)  # 'n d -> d n'

        encoding_indices = torch.argmin(d, dim=1)

        z_q = self.embedding(encoding_indices).view(z.shape)

        encodings = F.one_hot(encoding_indices, self.num_tokens).type(z.dtype)

        if not self.training:
            with torch.no_grad():
                cluster_size = encodings.sum(0)
                self.all_reduce_fn(cluster_size)
                ema_inplace(self.cluster_size, cluster_size, self.decay)

        if self.training and self.embedding.update:
            # EMA cluster size

            bins = encodings.sum(0)
            self.all_reduce_fn(bins)

            # self.embedding.cluster_size_ema_update(bins)
            ema_inplace(self.cluster_size, bins, self.decay)

            zero_mask = (bins == 0)
            bins = bins.masked_fill(zero_mask, 1.)

            embed_sum = z_flattened.t() @ encodings
            self.all_reduce_fn(embed_sum)

            embed_normalized = (embed_sum / bins.unsqueeze(0)).t()
            embed_normalized = l2norm(embed_normalized)

            embed_normalized = torch.where(zero_mask[..., None], self.embedding.weight,
                                           embed_normalized)
            norm_ema_inplace(self.embedding.weight, embed_normalized, self.decay)

        # compute loss for embedding
        loss = self.beta * F.mse_loss(z_q.detach(), z)

        # preserve gradients
        z_q = z + (z_q - z).detach()

        # reshape back to match original input shape
        # z_q, 'b h w c -> b c h w'
        # z_q = rearrange(z_q, 'b h w c -> b c h w')
        # z_q = z_q.transpose(1, 2)
        return z_q, loss, encoding_indices