from abc import abstractmethod import math import numpy as np import torch import torch as th import torch.nn as nn import torch.nn.functional as F import torchvision # For debugging, not actually used. from x_transformers.x_transformers import RelativePositionBias from models.audio.music.gpt_music import GptMusicLower from models.audio.music.music_quantizer import MusicQuantizer from models.diffusion.fp16_util import convert_module_to_f16, convert_module_to_f32 from models.diffusion.nn import ( conv_nd, linear, avg_pool_nd, zero_module, normalization, timestep_embedding, ) from models.lucidrains.x_transformers import Encoder from trainer.networks import register_model from utils.util import checkpoint, print_network, ceil_multiple class TimestepBlock(nn.Module): """ Any module where forward() takes timestep embeddings as a second argument. """ @abstractmethod def forward(self, x, emb): """ Apply the module to `x` given `emb` timestep embeddings. """ class TimestepEmbedSequential(nn.Sequential, TimestepBlock): """ A sequential module that passes timestep embeddings to the children that support it as an extra input. """ def forward(self, x, emb): for layer in self: if isinstance(layer, TimestepBlock): x = layer(x, emb) else: x = layer(x) return x class Upsample(nn.Module): """ An upsampling layer with an optional convolution. :param channels: channels in the inputs and outputs. :param use_conv: a bool determining if a convolution is applied. :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then upsampling occurs in the inner-two dimensions. """ def __init__(self, channels, use_conv, dims=2, out_channels=None, factor=None, ksize=3, pad=1): super().__init__() self.channels = channels self.out_channels = out_channels or channels self.use_conv = use_conv self.dims = dims if factor is None: if dims == 1: self.factor = 4 else: self.factor = 2 else: self.factor = factor if use_conv: if dims == 1: ksize = 5 pad = 2 self.conv = conv_nd(dims, self.channels, self.out_channels, ksize, padding=pad) def forward(self, x): assert x.shape[1] == self.channels if self.dims == 3: x = F.interpolate( x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2), mode="nearest" ) x = F.interpolate(x, scale_factor=self.factor, mode="nearest") if self.use_conv: x = self.conv(x) return x class Downsample(nn.Module): """ A downsampling layer with an optional convolution. :param channels: channels in the inputs and outputs. :param use_conv: a bool determining if a convolution is applied. :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then downsampling occurs in the inner-two dimensions. """ def __init__(self, channels, use_conv, dims=2, out_channels=None, factor=None, ksize=None, pad=None): super().__init__() self.channels = channels self.out_channels = out_channels or channels self.use_conv = use_conv self.dims = dims if ksize is None: ksize = 3 pad = 1 if dims == 1: ksize = 5 pad = 2 if dims == 1: stride = 4 elif dims == 2: stride = 2 else: stride = (1,2,2) if factor is not None: stride = factor if use_conv: self.op = conv_nd( dims, self.channels, self.out_channels, ksize, stride=stride, padding=pad ) else: assert self.channels == self.out_channels self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride) def forward(self, x): assert x.shape[1] == self.channels return self.op(x) class ResBlock(TimestepBlock): """ A residual block that can optionally change the number of channels. :param channels: the number of input channels. :param emb_channels: the number of timestep embedding channels. :param dropout: the rate of dropout. :param out_channels: if specified, the number of out channels. :param use_conv: if True and out_channels is specified, use a spatial convolution instead of a smaller 1x1 convolution to change the channels in the skip connection. :param dims: determines if the signal is 1D, 2D, or 3D. :param up: if True, use this block for upsampling. :param down: if True, use this block for downsampling. """ def __init__( self, channels, emb_channels, dropout, out_channels=None, use_conv=False, use_scale_shift_norm=False, dims=2, up=False, down=False, kernel_size=3, ): super().__init__() self.channels = channels self.emb_channels = emb_channels self.dropout = dropout self.out_channels = out_channels or channels self.use_conv = use_conv self.use_scale_shift_norm = use_scale_shift_norm padding = 1 if kernel_size == 3 else (2 if kernel_size == 5 else 0) self.in_layers = nn.Sequential( normalization(channels), nn.SiLU(), conv_nd(dims, channels, self.out_channels, kernel_size, padding=padding), ) self.updown = up or down if up: self.h_upd = Upsample(channels, False, dims) self.x_upd = Upsample(channels, False, dims) elif down: self.h_upd = Downsample(channels, False, dims) self.x_upd = Downsample(channels, False, dims) else: self.h_upd = self.x_upd = nn.Identity() self.emb_layers = nn.Sequential( nn.SiLU(), linear( emb_channels, 2 * self.out_channels if use_scale_shift_norm else self.out_channels, ), ) self.out_layers = nn.Sequential( normalization(self.out_channels), nn.SiLU(), nn.Dropout(p=dropout), zero_module( conv_nd(dims, self.out_channels, self.out_channels, kernel_size, padding=padding) ), ) if self.out_channels == channels: self.skip_connection = nn.Identity() elif use_conv: self.skip_connection = conv_nd( dims, channels, self.out_channels, kernel_size, padding=padding ) else: self.skip_connection = conv_nd(dims, channels, self.out_channels, 1) def forward(self, x, emb): """ Apply the block to a Tensor, conditioned on a timestep embedding. :param x: an [N x C x ...] Tensor of features. :param emb: an [N x emb_channels] Tensor of timestep embeddings. :return: an [N x C x ...] Tensor of outputs. """ return checkpoint( self._forward, x, emb ) def _forward(self, x, emb): if self.updown: in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1] h = in_rest(x) h = self.h_upd(h) x = self.x_upd(x) h = in_conv(h) else: h = self.in_layers(x) emb_out = self.emb_layers(emb).type(h.dtype) while len(emb_out.shape) < len(h.shape): emb_out = emb_out[..., None] if self.use_scale_shift_norm: out_norm, out_rest = self.out_layers[0], self.out_layers[1:] scale, shift = th.chunk(emb_out, 2, dim=1) h = out_norm(h) * (1 + scale) + shift h = out_rest(h) else: h = h + emb_out h = self.out_layers(h) return self.skip_connection(x) + h class AttentionBlock(nn.Module): """ An attention block that allows spatial positions to attend to each other. Originally ported from here, but adapted to the N-d case. https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/models/unet.py#L66. """ def __init__( self, channels, num_heads=1, num_head_channels=-1, use_new_attention_order=False, do_checkpoint=True, relative_pos_embeddings=False, ): super().__init__() self.channels = channels self.do_checkpoint = do_checkpoint if num_head_channels == -1: self.num_heads = num_heads else: assert ( channels % num_head_channels == 0 ), f"q,k,v channels {channels} is not divisible by num_head_channels {num_head_channels}" self.num_heads = channels // num_head_channels self.norm = normalization(channels) self.qkv = conv_nd(1, channels, channels * 3, 1) if use_new_attention_order: # split qkv before split heads self.attention = QKVAttention(self.num_heads) else: # split heads before split qkv self.attention = QKVAttentionLegacy(self.num_heads) self.proj_out = zero_module(conv_nd(1, channels, channels, 1)) if relative_pos_embeddings: self.relative_pos_embeddings = RelativePositionBias(scale=(channels // self.num_heads) ** .5, causal=False, heads=num_heads, num_buckets=32, max_distance=64) else: self.relative_pos_embeddings = None def forward(self, x, mask=None): if self.do_checkpoint: return checkpoint(self._forward, x, mask) else: return self._forward(x, mask) def _forward(self, x, mask): b, c, *spatial = x.shape x = x.reshape(b, c, -1) qkv = self.qkv(self.norm(x)) h = self.attention(qkv, mask, self.relative_pos_embeddings) h = self.proj_out(h) return (x + h).reshape(b, c, *spatial) def count_flops_attn(model, _x, y): """ A counter for the `thop` package to count the operations in an attention operation. Meant to be used like: macs, params = thop.profile( model, inputs=(inputs, timestamps), custom_ops={QKVAttention: QKVAttention.count_flops}, ) """ b, c, *spatial = y[0].shape num_spatial = int(np.prod(spatial)) # We perform two matmuls with the same number of ops. # The first computes the weight matrix, the second computes # the combination of the value vectors. matmul_ops = 2 * b * (num_spatial ** 2) * c model.total_ops += th.DoubleTensor([matmul_ops]) class QKVAttentionLegacy(nn.Module): """ A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping """ def __init__(self, n_heads): super().__init__() self.n_heads = n_heads def forward(self, qkv, mask=None, rel_pos=None): """ Apply QKV attention. :param qkv: an [N x (H * 3 * C) x T] tensor of Qs, Ks, and Vs. :return: an [N x (H * C) x T] tensor after attention. """ bs, width, length = qkv.shape assert width % (3 * self.n_heads) == 0 ch = width // (3 * self.n_heads) q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch, dim=1) scale = 1 / math.sqrt(math.sqrt(ch)) weight = th.einsum( "bct,bcs->bts", q * scale, k * scale ) # More stable with f16 than dividing afterwards if rel_pos is not None: weight = rel_pos(weight.reshape(bs, self.n_heads, weight.shape[-2], weight.shape[-1])).reshape(bs * self.n_heads, weight.shape[-2], weight.shape[-1]) weight = th.softmax(weight.float(), dim=-1).type(weight.dtype) if mask is not None: # The proper way to do this is to mask before the softmax using -inf, but that doesn't work properly on CPUs. mask = mask.repeat(self.n_heads, 1).unsqueeze(1) weight = weight * mask a = th.einsum("bts,bcs->bct", weight, v) return a.reshape(bs, -1, length) @staticmethod def count_flops(model, _x, y): return count_flops_attn(model, _x, y) class QKVAttention(nn.Module): """ A module which performs QKV attention and splits in a different order. """ def __init__(self, n_heads): super().__init__() self.n_heads = n_heads def forward(self, qkv, mask=None, rel_pos=None): """ Apply QKV attention. :param qkv: an [N x (3 * H * C) x T] tensor of Qs, Ks, and Vs. :return: an [N x (H * C) x T] tensor after attention. """ bs, width, length = qkv.shape assert width % (3 * self.n_heads) == 0 ch = width // (3 * self.n_heads) q, k, v = qkv.chunk(3, dim=1) scale = 1 / math.sqrt(math.sqrt(ch)) weight = th.einsum( "bct,bcs->bts", (q * scale).view(bs * self.n_heads, ch, length), (k * scale).view(bs * self.n_heads, ch, length), ) # More stable with f16 than dividing afterwards if rel_pos is not None: weight = rel_pos(weight) if mask is not None: # The proper way to do this is to mask before the softmax using -inf, but that doesn't work properly on CPUs. mask = mask.repeat(self.n_heads, 1).unsqueeze(1) weight = weight * mask weight = th.softmax(weight.float(), dim=-1).type(weight.dtype) a = th.einsum("bts,bcs->bct", weight, v.reshape(bs * self.n_heads, ch, length)) return a.reshape(bs, -1, length) @staticmethod def count_flops(model, _x, y): return count_flops_attn(model, _x, y) class UNetMusicModel(nn.Module): """ The full UNet model with attention and timestep embedding. :param in_channels: channels in the input Tensor. :param model_channels: base channel count for the model. :param out_channels: channels in the output Tensor. :param num_res_blocks: number of residual blocks per downsample. :param attention_resolutions: a collection of downsample rates at which attention will take place. May be a set, list, or tuple. For example, if this contains 4, then at 4x downsampling, attention will be used. :param dropout: the dropout probability. :param channel_mult: channel multiplier for each level of the UNet. :param conv_resample: if True, use learned convolutions for upsampling and downsampling. :param dims: determines if the signal is 1D, 2D, or 3D. :param num_classes: if specified (as an int), then this model will be class-conditional with `num_classes` classes. :param num_heads: the number of attention heads in each attention layer. :param num_heads_channels: if specified, ignore num_heads and instead use a fixed channel width per attention head. :param num_heads_upsample: works with num_heads to set a different number of heads for upsampling. Deprecated. :param use_scale_shift_norm: use a FiLM-like conditioning mechanism. :param resblock_updown: use residual blocks for up/downsampling. :param use_new_attention_order: use a different attention pattern for potentially increased efficiency. """ def __init__( self, in_channels, input_vec_dim, model_channels, out_channels, num_res_blocks, attention_resolutions, dropout=0, channel_mult=(1, 2, 4, 8), ar_prior=False, conv_resample=True, dims=2, num_classes=None, use_fp16=False, num_heads=1, num_head_channels=-1, num_heads_upsample=-1, use_scale_shift_norm=False, resblock_updown=False, use_new_attention_order=False, use_raw_y_as_embedding=False, unconditioned_percentage=.1, # This implements a mechanism similar to what is used in classifier-free training. ): super().__init__() if num_heads_upsample == -1: num_heads_upsample = num_heads self.in_channels = in_channels self.model_channels = model_channels self.out_channels = out_channels self.num_res_blocks = num_res_blocks self.attention_resolutions = attention_resolutions self.dropout = dropout self.channel_mult = channel_mult self.conv_resample = conv_resample self.num_classes = num_classes self.dtype = th.float16 if use_fp16 else th.float32 self.num_heads = num_heads self.num_head_channels = num_head_channels self.num_heads_upsample = num_heads_upsample self.unconditioned_percentage = unconditioned_percentage self.ar_prior = ar_prior time_embed_dim = model_channels * 4 self.time_embed = nn.Sequential( linear(model_channels, time_embed_dim), nn.SiLU(), linear(time_embed_dim, time_embed_dim), ) if self.ar_prior: self.ar_input = nn.Linear(input_vec_dim, model_channels) self.ar_prior_intg = Encoder( dim=model_channels, depth=4, heads=num_heads, ff_dropout=dropout, attn_dropout=dropout, use_rmsnorm=True, ff_glu=True, rotary_pos_emb=True, zero_init_branch_output=True, ff_mult=1, ) else: self.input_converter = nn.Linear(input_vec_dim, model_channels) self.code_converter = Encoder( dim=model_channels, depth=4, heads=num_heads, ff_dropout=dropout, attn_dropout=dropout, use_rmsnorm=True, ff_glu=True, rotary_pos_emb=True, zero_init_branch_output=True, ff_mult=1, ) self.unconditioned_embedding = nn.Parameter(torch.randn(1,1,model_channels)) self.x_processor = conv_nd(dims, in_channels, model_channels, 3, padding=1) if self.num_classes is not None: self.label_emb = nn.Embedding(num_classes, time_embed_dim) self.use_raw_y_as_embedding = use_raw_y_as_embedding assert not ((self.num_classes is not None) and use_raw_y_as_embedding) # These are mutually-exclusive. self.input_blocks = nn.ModuleList( [ TimestepEmbedSequential( conv_nd(dims, model_channels*2, model_channels, 1, padding=0) ) ] ) self._feature_size = model_channels input_block_chans = [model_channels] ch = model_channels ds = 1 for level, mult in enumerate(channel_mult): for _ in range(num_res_blocks): layers = [ ResBlock( ch, time_embed_dim, dropout, out_channels=int(mult * model_channels), dims=dims, use_scale_shift_norm=use_scale_shift_norm, ) ] ch = int(mult * model_channels) if ds in attention_resolutions: layers.append( AttentionBlock( ch, num_heads=num_heads, num_head_channels=num_head_channels, use_new_attention_order=use_new_attention_order, ) ) self.input_blocks.append(TimestepEmbedSequential(*layers)) self._feature_size += ch input_block_chans.append(ch) if level != len(channel_mult) - 1: out_ch = ch self.input_blocks.append( TimestepEmbedSequential( ResBlock( ch, time_embed_dim, dropout, out_channels=out_ch, dims=dims, use_scale_shift_norm=use_scale_shift_norm, down=True, ) if resblock_updown else Downsample( ch, conv_resample, dims=dims, out_channels=out_ch ) ) ) ch = out_ch input_block_chans.append(ch) ds *= 2 self._feature_size += ch self.middle_block = TimestepEmbedSequential( ResBlock( ch, time_embed_dim, dropout, dims=dims, use_scale_shift_norm=use_scale_shift_norm, ), AttentionBlock( ch, num_heads=num_heads, num_head_channels=num_head_channels, use_new_attention_order=use_new_attention_order, ), ResBlock( ch, time_embed_dim, dropout, dims=dims, use_scale_shift_norm=use_scale_shift_norm, ), ) self._feature_size += ch self.output_blocks = nn.ModuleList([]) for level, mult in list(enumerate(channel_mult))[::-1]: for i in range(num_res_blocks + 1): ich = input_block_chans.pop() layers = [ ResBlock( ch + ich, time_embed_dim, dropout, out_channels=int(model_channels * mult), dims=dims, use_scale_shift_norm=use_scale_shift_norm, ) ] ch = int(model_channels * mult) if ds in attention_resolutions: layers.append( AttentionBlock( ch, num_heads=num_heads_upsample, num_head_channels=num_head_channels, use_new_attention_order=use_new_attention_order, ) ) if level and i == num_res_blocks: out_ch = ch layers.append( ResBlock( ch, time_embed_dim, dropout, out_channels=out_ch, dims=dims, use_scale_shift_norm=use_scale_shift_norm, up=True, ) if resblock_updown else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch) ) ds //= 2 self.output_blocks.append(TimestepEmbedSequential(*layers)) self._feature_size += ch self.out = nn.Sequential( normalization(ch), nn.SiLU(), zero_module(conv_nd(dims, model_channels, out_channels, 3, padding=1)), ) def forward(self, x, timesteps, y, conditioning_free=False): orig_x_shape = x.shape[-1] cm = ceil_multiple(x.shape[-1], 16) if cm != 0: pc = (cm - x.shape[-1]) / x.shape[-1] x = F.pad(x, (0, cm - x.shape[-1])) y = F.pad(y.permute(0,2,1), (0, int(pc * y.shape[-1]))).permute(0,2,1) unused_params = [] hs = [] emb = self.time_embed(timestep_embedding(timesteps, self.model_channels)) if conditioning_free: expanded_code_emb = self.unconditioned_embedding.repeat(x.shape[0], x.shape[-1], 1).permute(0,2,1) if self.ar_prior: unused_params.extend(list(self.ar_input.parameters()) + list(self.ar_prior_intg.parameters())) else: unused_params.extend(list(self.input_converter.parameters()) + list(self.code_converter.parameters())) else: code_emb = self.ar_input(y) if self.ar_prior else self.input_converter(y) if self.training and self.unconditioned_percentage > 0: unconditioned_batches = torch.rand((code_emb.shape[0], 1, 1), device=code_emb.device) < self.unconditioned_percentage code_emb = torch.where(unconditioned_batches, self.unconditioned_embedding.repeat(y.shape[0], 1, 1), code_emb) code_emb = self.ar_prior_intg(code_emb) if self.ar_prior else self.code_converter(code_emb) expanded_code_emb = F.interpolate(code_emb.permute(0,2,1), size=x.shape[-1], mode='nearest') h = x.type(self.dtype) expanded_code_emb = expanded_code_emb.type(self.dtype) h = self.x_processor(h) h = torch.cat([h, expanded_code_emb], dim=1) for module in self.input_blocks: h = module(h, emb) hs.append(h) h = self.middle_block(h, emb) for module in self.output_blocks: h = th.cat([h, hs.pop()], dim=1) h = module(h, emb) h = h.type(x.dtype) # Involve probabilistic or possibly unused parameters in loss so we don't get DDP errors. extraneous_addition = 0 for p in unused_params: extraneous_addition = extraneous_addition + p.mean() h = h + extraneous_addition * 0 out = self.out(h) return out[:, :, :orig_x_shape] class UNetMusicModelWithQuantizer(nn.Module): def __init__(self, freeze_quantizer_until=20000, **kwargs): super().__init__() self.internal_step = 0 self.freeze_quantizer_until = freeze_quantizer_until self.diff = UNetMusicModel(**kwargs) self.m2v = MusicQuantizer(inp_channels=256, inner_dim=[1024,1024,512], codevector_dim=1024, codebook_size=512, codebook_groups=2) self.m2v.quantizer.temperature = self.m2v.min_gumbel_temperature del self.m2v.up def update_for_step(self, step, *args): self.internal_step = step qstep = max(0, self.internal_step - self.freeze_quantizer_until) self.m2v.quantizer.temperature = max( self.m2v.max_gumbel_temperature * self.m2v.gumbel_temperature_decay**qstep, self.m2v.min_gumbel_temperature, ) def forward(self, x, timesteps, truth_mel, disable_diversity=False, conditioning_input=None, conditioning_free=False): quant_grad_enabled = self.internal_step > self.freeze_quantizer_until with torch.set_grad_enabled(quant_grad_enabled): proj, diversity_loss = self.m2v(truth_mel, return_decoder_latent=True) proj = proj.permute(0,2,1) # Make sure this does not cause issues in DDP by explicitly using the parameters for nothing. if not quant_grad_enabled: unused = 0 for p in self.m2v.parameters(): unused = unused + p.mean() * 0 proj = proj + unused diversity_loss = diversity_loss * 0 diff = self.diff(x, timesteps, proj, conditioning_free=conditioning_free) if disable_diversity: return diff return diff, diversity_loss def get_debug_values(self, step, __): if self.m2v.total_codes > 0: return {'histogram_codes': self.m2v.codes[:self.m2v.total_codes]} else: return {} class UNetMusicModelARPrior(nn.Module): def __init__(self, freeze_unet=False, **kwargs): super().__init__() self.internal_step = 0 self.ar = GptMusicLower(dim=512, layers=12) for p in self.ar.parameters(): p.DO_NOT_TRAIN = True p.requires_grad = False self.diff = UNetMusicModel(ar_prior=True, **kwargs) if freeze_unet: for p in self.diff.parameters(): p.DO_NOT_TRAIN = True p.requires_grad = False for p in list(self.diff.ar_prior_intg.parameters()) + list(self.diff.ar_input.parameters()): del p.DO_NOT_TRAIN p.requires_grad = True def get_grad_norm_parameter_groups(self): groups = { 'input_blocks': list(self.diff.input_blocks.parameters()), 'output_blocks': list(self.diff.output_blocks.parameters()), 'ar_prior_intg': list(self.diff.ar_prior_intg.parameters()), 'time_embed': list(self.diff.time_embed.parameters()), } return groups def forward(self, x, timesteps, truth_mel, disable_diversity=False, conditioning_input=None, conditioning_free=False): with torch.no_grad(): prior = self.ar(truth_mel, conditioning_input, return_latent=True) diff = self.diff(x, timesteps, prior, conditioning_free=conditioning_free) return diff @register_model def register_unet_diffusion_music_codes(opt_net, opt): return UNetMusicModelWithQuantizer(**opt_net['args']) @register_model def register_unet_diffusion_music_ar_prior(opt_net, opt): return UNetMusicModelARPrior(**opt_net['args']) if __name__ == '__main__': clip = torch.randn(2, 256, 300) cond = torch.randn(2, 256, 300) ts = torch.LongTensor([600, 600]) model = UNetMusicModelARPrior(in_channels=256, out_channels=512, model_channels=640, num_res_blocks=3, input_vec_dim=512, attention_resolutions=(2,4), channel_mult=(1,2,3), dims=1, use_scale_shift_norm=True, dropout=.1, num_heads=8, unconditioned_percentage=.4, freeze_unet=True) print_network(model) model.get_grad_norm_parameter_groups() ar_weights = torch.load('D:\\dlas\\experiments\\train_music_gpt\\models\\44500_generator_ema.pth') model.ar.load_state_dict(ar_weights, strict=True) diff_weights = torch.load('X:\\dlas\\experiments\\train_music_diffusion_unet_music\\models\\55500_generator_ema.pth') pruned_diff_weights = {} for k,v in diff_weights.items(): if k.startswith('diff.'): pruned_diff_weights[k.replace('diff.', '')] = v model.diff.load_state_dict(pruned_diff_weights, strict=False) torch.save(model.state_dict(), 'sample.pth') model(clip, ts, cond, cond)