audiocraft/lm.py

from dataclasses import dataclass
import logging
import math
import typing as tp
import torch
import torch.nn.functional as F
from audiocraft.transformer import StreamingTransformer
from dataclasses import dataclass
from functools import partial
from torch import nn
from audiocraft.activations import get_activation_fn
import numpy as np

def _shift(x):
    # cyclic shift of [1, 4, seq_len] slices from [bs, 4, seq_len]
    print(x.shape, 'SHIFT\n= = = = = ')
    for i, _slice in enumerate(x):
        n = x.shape[2]
        offset = np.random.randint(.24 * n, max(1, .74 * n))  # high should be above >= 0 TBD
        print(offset)
        x[i, :, :] = torch.roll(_slice, offset, dims=1)
    return x



def get_init_fn(method: str, input_dim: int, init_depth: tp.Optional[int] = None):
    """LM layer initialization.
    Inspired from xlformers: https://github.com/fairinternal/xlformers

    Args:
        method (str): Method name for init function. Valid options are:
            'gaussian', 'uniform'.
        input_dim (int): Input dimension of the initialized module.
        init_depth (int, optional): Optional init depth value used to rescale
            the standard deviation if defined.
    """
    # Compute std
    std = 1 / math.sqrt(input_dim)
    # Rescale with depth
    if init_depth is not None:
        std = std / math.sqrt(2 * init_depth)

    if method == 'gaussian':
        return partial(
            torch.nn.init.trunc_normal_, mean=0.0, std=std, a=-3 * std, b=3 * std
        )
    elif method == 'uniform':
        bound = math.sqrt(3) * std  # ensure the standard deviation is `std`
        return partial(torch.nn.init.uniform_, a=-bound, b=bound)
    else:
        raise ValueError("Unsupported layer initialization method")


def init_layer(m: nn.Module,
               method: str,
               init_depth: tp.Optional[int] = None,
               zero_bias_init: bool = False):
    """Wrapper around ``get_init_fn`` for proper initialization of LM modules.

    Args:
        m (nn.Module): Module to initialize.
        method (str): Method name for the init function.
        init_depth (int, optional): Optional init depth value used to rescale
            the standard deviation if defined.
        zero_bias_init (bool): Whether to initialize the bias to 0 or not.
    """
    if isinstance(m, nn.Linear):
        init_fn = get_init_fn(method, m.in_features, init_depth=init_depth)
        if m.weight.device.type == 'cpu' and m.weight.dtype == torch.float16:
            weight = m.weight.float()
            init_fn(weight)
            m.weight.data[:] = weight.half()
        else:
            init_fn(m.weight)
        if zero_bias_init and m.bias is not None:
            nn.init.constant_(m.bias, 0)
    elif isinstance(m, nn.Embedding):
        init_fn = get_init_fn(method, m.embedding_dim, init_depth=None)
        if m.weight.device.type == 'cpu' and m.weight.dtype == torch.float16:
            weight = m.weight.float()
            init_fn(weight)
            m.weight.data[:] = weight.half()
        else:
            init_fn(m.weight)


class ScaledEmbedding(nn.Embedding):
    """Boost learning rate for embeddings (with `scale`).
    """
    def __init__(self, *args, lr=None, **kwargs):
        super().__init__(*args, **kwargs)
        self.lr = lr

    def make_optim_group(self):
        group = {"params": list(self.parameters())}
        if self.lr is not None:
            group["lr"] = self.lr
        return group


@dataclass
class LMOutput:
    # The logits are already re-aligned with the input codes
    # hence no extra shift is required, e.g. when computing CE
    logits: torch.Tensor  # [B, K, T, card]
    mask: torch.Tensor  # [B, K, T]


class LMModel(nn.Module):
    """Transformer-based language model on multiple streams of codes.

    Args:
        pattern_provider (CodebooksPatternProvider): Pattern provider for codebook interleaving.
        condition_provider (MusicConditioningProvider): Conditioning provider from metadata.
        fuser (ConditionFuser): Fuser handling the fusing of conditions with language model input.
        n_q (int): Number of parallel streams to model.
        card (int): Cardinality, vocabulary size.
        dim (int): Dimension of the transformer encoder.
        num_heads (int): Number of heads for the transformer encoder.
        hidden_scale (int): Scale for hidden feed forward dimension of the transformer encoder.
        norm (str): Normalization method.
        norm_first (bool): Use pre-norm instead of post-norm.
        emb_lr (float, optional): Embedding-specific learning rate.
        bias_proj (bool): Use bias for output projections.
        weight_init (str, optional): Method for weight initialization.
        depthwise_init (str, optional): Method for depthwise weight initialization.
        zero_bias_init (bool): If true and bias in Linears, initialize bias to zeros.
        cfg_dropout (float): Classifier-free guidance dropout.
        cfg_coef (float): Classifier-free guidance coefficient.
        attribute_dropout (dict): Attribute dropout probabilities.
        two_step_cfg (bool): Whether to run classifier free-guidance with 2 distinct steps.
        **kwargs: Additional parameters for the transformer encoder.
    """
    def __init__(self,
                 pattern_provider, 
                 condition_provider,
                 n_q: int = 8, 
                 card: int = 1024, 
                 dim: int = 128,
                 num_heads: int = 8,
                 hidden_scale: int = 4, 
                 norm: str = 'layer_norm',
                 norm_first: bool = False,
                 emb_lr: tp.Optional[float] = None, 
                 bias_proj: bool = True,
                 weight_init: tp.Optional[str] = None, 
                 depthwise_init: tp.Optional[str] = None,
                 zero_bias_init: bool = False, cfg_dropout: float = 0, 
                 cfg_coef: float = 1.0,
                 two_step_cfg: bool = False,
                 **kwargs):
        super().__init__()
        self.cfg_coef = cfg_coef
        self.condition_provider = condition_provider
        self.card = card  # 2048 ?
        self.n_draw = 1  # replicate so many times the generation of each text in batch
        embed_dim = self.card + 1
        self.n_q = n_q
        self.dim = dim
        self.pattern_provider = pattern_provider
        self.two_step_cfg = two_step_cfg
        self.emb = nn.ModuleList([ScaledEmbedding(embed_dim, dim, lr=emb_lr) for _ in range(n_q)])
        if 'activation' in kwargs:
            kwargs['activation'] = get_activation_fn(kwargs['activation'])
        # ========================================================================
        #  {
        #   'dtype': torch.float16, 'device': 'cuda',
        #   'num_layers': 48, 'dropout': 0.0, 'activation': 'gelu', 
        #   'bias_ff': False, 'bias_attn': False,
        #   'past_context': None, 'causal': True, 
        #   'custom': False, 'memory_efficient': True,
        #   'attention_as_float32': False, 'positional_embedding': 'sin', 'xpos': False, 
        #   'checkpointing': 'none', 'cross_attention': True, 'qk_layer_norm': False,
        #   'qk_layer_norm_cross': False, 'attention_dropout': None, 'kv_repeat': 1
        #   }
        # ==========================================================================
        kwargs.pop('layer_scale')  # nn.Indentity()
        
        self.transformer = StreamingTransformer(
            d_model=dim, 
            num_heads=num_heads, 
            dim_feedforward=int(hidden_scale * dim),
            norm=norm, 
            norm_first=norm_first, **kwargs)
        self.out_norm: tp.Optional[nn.Module] = None
        if norm_first:
            self.out_norm = nn.LayerNorm(dim, eps=1e-5)
        self.linears = nn.ModuleList([nn.Linear(dim, self.card, bias=bias_proj) for _ in range(n_q)])
        self._init_weights(weight_init, depthwise_init, zero_bias_init)
        self._fsdp: tp.Optional[nn.Module]
        self.__dict__['_fsdp'] = None

    def _init_weights(self, weight_init: tp.Optional[str], depthwise_init: tp.Optional[str], zero_bias_init: bool):
        """Initialization of the transformer module weights.

        Args:
            weight_init (str, optional): Weight initialization strategy. See ``get_init_fn`` for valid options.
            depthwise_init (str, optional): Depthwise initialization strategy. The following options are valid:
                'current' where the depth corresponds to the current layer index or 'global' where the total number
                of layer is used as depth. If not set, no depthwise initialization strategy is used.
            zero_bias_init (bool): Whether to initialize bias to zero or not.
        """
        assert depthwise_init is None or depthwise_init in ['current', 'global']
        assert depthwise_init is None or weight_init is not None, \
            "If 'depthwise_init' is defined, a 'weight_init' method should be provided."
        assert not zero_bias_init or weight_init is not None, \
            "If 'zero_bias_init', a 'weight_init' method should be provided"

        if weight_init is None:
            return

        for emb_layer in self.emb:
            init_layer(emb_layer, method=weight_init, init_depth=None, zero_bias_init=zero_bias_init)

        for layer_idx, tr_layer in enumerate(self.transformer.layers):
            depth = None
            if depthwise_init == 'current':
                depth = layer_idx + 1
            elif depthwise_init == 'global':
                depth = len(self.transformer.layers)
            init_fn = partial(init_layer,
                              method=weight_init,
                              init_depth=depth, 
                              zero_bias_init=zero_bias_init)
            tr_layer.apply(init_fn)

        for linear in self.linears:
            init_layer(linear, method=weight_init, init_depth=None, zero_bias_init=zero_bias_init)

    @property
    def special_token_id(self) -> int:
        return self.card
    
        

    def forward(self,
                sequence,
                condition_tensors=None,
                token_count=None):
        # takes bs=3 duplicates null condition to bs=6 splits logits to cfg returns bs=3 

        bs, _, _ = sequence.shape         # sequence [bs, n_draw,4]

        input_ = sum([self.emb[k](sequence[:, k]) for k in range(self.n_q)])
        out = self.transformer(torch.cat([input_, input_], 0),
                               cross_attention_src=condition_tensors,
                               token_count=token_count)
        if self.out_norm:
            out = self.out_norm(out)

        logits = torch.stack([self.linears[k](out) for k in range(self.n_q)], dim=1)#[2*bs,4,1,2048]
        
        logits = 3 * logits[:bs, :, :, :] - 2 * logits[bs:, :, :, :]  # [3, 4, 1, 2048]
        
        # SAMPLE TOP K
        k = 250
        p = torch.softmax(logits, dim=3)
        top_k_value, _ = torch.topk(p, k, dim=3)  # [3, 4, 1, k]
        min_value_top_k = top_k_value[:, :, :, -1:] 
        p *= (p >= min_value_top_k).float()   # zero low probs
        p.div_(p.sum(dim=-1, keepdim=True))   # renormalise on non-zero probs

        
        # BRING THE nq = 4 IN BATCH
        p = p.reshape(bs * self.n_q, 2048)
        out = torch.multinomial(p,  # p=[bs,2048], out=[bs, num_samples]
                                num_samples=self.n_draw,
                                replacement=True)  # [bs*4, self.n_draw]
        return out.reshape(bs, self.n_q, self.n_draw).transpose(1,2)  # [bs=3not6, self.n_draw, 4]

    @torch.no_grad()
    def generate(self,
                 descriptions = ['windy day', 'rain storm'],
                 max_gen_len = 256):
        
        text_condition = self.condition_provider(descriptions)
                
        # NULL CONDITION
        # text_condition = cfg_conditions['description'][0]
        bs, _, _ = text_condition.shape
        text_condition = torch.cat(
            [
                text_condition,
                torch.zeros_like(text_condition)
            ], 0)
        
        

        
        pattern = self.pattern_provider.get_pattern(max_gen_len)
        gen_codes = torch.full((bs, 
                                self.n_q, 
                                max_gen_len), -1, dtype=torch.long, 
                                device=text_condition.device)

        gen_sequence, _, mask = pattern.build_pattern_sequence(gen_codes, self.special_token_id)
        _, _, audiodur = gen_sequence.shape  # bs, 4, 7=audiodur
        
        # print(gen_sequence.shape, mask.shape, 'F')  # mask has no batch = [4,audio_duration]
        # print(f'{mask=}')
        # 
        # torch.Size([3, 4, 7]) torch.Size([4, 7]) F
        # mask=tensor([[False,  True,  True,  True, False, False, False],
        #              [False, False,  True,  True,  True, False, False],
        #              [False, False, False,  True,  True,  True, False],
        #              [False, False, False, False,  True,  True,  True]], device='cuda:0')

        mask = mask[None, None, :, :].repeat(bs, self.n_draw, 1, 1)  # [bs, n_draw, 4, audio duration]
        gen_sequence = gen_sequence[:, None, :, :].repeat(1, self.n_draw, 1, 1)  # bs,n_draw,4,dur
        

        
        for offset in range(1, audiodur):
            
            # forward duplicates the query to nullcond - then cfg & returns deduplicate token
            next_token = self.forward(gen_sequence[:, 0, :, offset-1:offset],
                                      condition_tensors=text_condition,  # utilisation of the attention mask of txt condition ?
                                      token_count=offset-1)  # [bs, 4, 1, 2048]
            


            
            # MASK is not full 1---- HAS 4 x audioduration PATTERN
            m = mask[:, :, :, offset]
            next_token[~m] = self.special_token_id
            gen_sequence[:, :, :, offset] = torch.where(
                gen_sequence[:, :, :, offset] == -1, #unknown_token,
                next_token,
                gen_sequence[:, :, :, offset]
            )
            

        # 1. reshape n_draw as bs * n_draw
        # 2. invert all short-sequences
        # 3. reshape bs * n_draw -> bs, n_draw * audiodur ELONGATION 
        out_codes, _, _ = pattern.revert_pattern_sequence(
            gen_sequence.reshape(bs * self.n_draw, 4, audiodur),  # [3,8,4,7]
            special_token=-1)
        # print(f'{gen_sequence.shape=} {out_codes.shape=} Ha')  # REVERT PATTERN REDUCES DURATION?
        _, _, new_len = out_codes.shape                        # 4 IS PRESERVED AFTER REVERT!
        out_codes = out_codes.reshape(bs, self.n_draw, 4, new_len)
        out_codes = out_codes.transpose(1, 2).reshape(bs, 4, self.n_draw * new_len) 
        print(out_codes.shape, 'o')
        for _ in range(7):
            out_codes = _shift(out_codes)
        
        # Clear Transformer k/v history (Different history is kept by 48x selfattn)
        for lay in self.transformer.layers:
             lay.self_attn.k_history = None
             lay.self_attn.v_history = None

        return out_codes