Checkin

codes/models/audio/music/transformer_diffusion12.py

@@ -609,7 +609,6 @@ def test_vqvae_model():
     o = model(clip, ts, cond)
     pg = model.get_grad_norm_parameter_groups()
 
-    """
     with torch.no_grad():
         proj = torch.randn(2, 100, 512).cuda()
         clip = clip.cuda()
@@ -618,10 +617,9 @@ def test_vqvae_model():
         model = model.cuda().eval()
         model.diff.enable_fp16 = True
         ti = model.diff.timestep_independent(proj, clip.shape[2])
-        for k in range(100):
+        for k in range(1000):
             model.diff(clip, ts, precomputed_code_embeddings=ti)
         print(f"Elapsed: {time()-start}")
-        """
 
 
 def test_multi_vqvae_model():
@@ -690,4 +688,5 @@ def extract_diff(in_f, out_f, remove_head=False):
 if __name__ == '__main__':
     #extract_diff('X:\\dlas\\experiments\\train_music_diffusion_tfd12\\models\\41000_generator_ema.pth', 'extracted_diff.pth', True)
     #test_cheater_model()
-    extract_diff('X:\\dlas\experiments\\train_music_diffusion_tfd_cheater_from_scratch\\models\\56500_generator_ema.pth', 'extracted.pth', remove_head=True)
+    test_vqvae_model()
+    #extract_diff('X:\\dlas\experiments\\train_music_diffusion_tfd_cheater_from_scratch\\models\\56500_generator_ema.pth', 'extracted.pth', remove_head=True)

codes/models/audio/music/transformer_diffusion13.py

@@ -19,8 +19,8 @@ class SubBlock(nn.Module):
         super().__init__()
         self.dropout = nn.Dropout(p=dropout)
         self.attn = AttentionBlock(inp_dim, out_channels=contraction_dim, num_heads=heads)
-        self.register_buffer('mask', build_local_attention_mask(n=4000, l=64), persistent=False)
+        self.register_buffer('mask', build_local_attention_mask(n=6000, l=64), persistent=False)
-        self.pos_bias = RelativeQKBias(l=64)
+        self.pos_bias = RelativeQKBias(l=64, max_positions=6000)
         ff_contract = contraction_dim//2
         self.ff1 = nn.Sequential(nn.Conv1d(inp_dim+contraction_dim, ff_contract, kernel_size=1),
                                  nn.GroupNorm(8, ff_contract),

codes/models/audio/tts/unet_diffusion_tts_flat.py

@@ -7,7 +7,8 @@ import torch.nn.functional as F
 from torch import autocast
 
 from models.diffusion.nn import timestep_embedding, normalization, zero_module, conv_nd, linear
-from models.diffusion.unet_diffusion import AttentionBlock, TimestepEmbedSequential, TimestepBlock
+from models.diffusion.unet_diffusion import TimestepEmbedSequential, TimestepBlock, QKVAttentionLegacy
+from models.lucidrains.x_transformers import RelativePositionBias
 from trainer.networks import register_model
 from utils.util import checkpoint
 
@@ -19,6 +20,52 @@ def is_sequence(t):
     return t.dtype == torch.long
 
 
+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,
+        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 = nn.Conv1d(channels, channels * 3, 1)
+        # split heads before split qkv
+        self.attention = QKVAttentionLegacy(self.num_heads)
+
+        self.proj_out = zero_module(nn.Conv1d(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):
+        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)
+
+
 class ResBlock(TimestepBlock):
     def __init__(
         self,
@@ -336,7 +383,7 @@ if __name__ == '__main__':
         start = time()
         model = model.cuda().eval()
         ti = model.timestep_independent(proj, clip, clip.shape[2], False)
-        for k in range(100):
+        for k in range(1000):
             model(clip, ts, precomputed_aligned_embeddings=ti)
         print(f"Elapsed: {time()-start}")
 

codes/trainer/eval/music_diffusion_fid.py

@@ -137,8 +137,8 @@ class MusicDiffusionFid(evaluator.Evaluator):
         #    s = q9.clamp(1, 9999999999)
         #    x = x.clamp(-s, s) / s
         #    return x
-        perp = self.diffuser.p_sample_loop_for_log_perplexity(self.model, mel_norm,
+        #perp = self.diffuser.p_sample_loop_for_log_perplexity(self.model, mel_norm,
-                                                              model_kwargs = {'truth_mel': mel_norm})
+        #                                                      model_kwargs = {'truth_mel': mel_norm})
 
         sampler = self.diffuser.ddim_sample_loop if self.ddim else self.diffuser.p_sample_loop
         gen_mel = sampler(self.model, mel_norm.shape, model_kwargs={'truth_mel': mel_norm})
@@ -155,7 +155,7 @@ class MusicDiffusionFid(evaluator.Evaluator):
                                               model_kwargs={'codes': mel})
         real_wav = pixel_shuffle_1d(real_wav, 16)
 
-        return gen_wav, real_wav.squeeze(0), gen_mel, mel_norm, sample_rate, perp
+        return gen_wav, real_wav.squeeze(0), gen_mel, mel_norm, sample_rate, torch.tensor([0])
 
     def perform_reconstruction_from_cheater_gen(self, audio, sample_rate=22050):
         audio = audio.unsqueeze(0)
@@ -303,7 +303,7 @@ class MusicDiffusionFid(evaluator.Evaluator):
             gen_projections = torch.stack(gen_projections, dim=0)
             real_projections = torch.stack(real_projections, dim=0)
             frechet_distance = torch.tensor(self.compute_frechet_distance(gen_projections, real_projections), device=self.env['device'])
-            perplexity = torch.stack(perplexities, dim=0).mean()
+            perplexity = torch.stack(perplexities, dim=0).float().mean()
 
             if distributed.is_initialized() and distributed.get_world_size() > 1:
                 distributed.all_reduce(frechet_distance)
@@ -338,16 +338,17 @@ if __name__ == '__main__':
     # For TFD+cheater trainer
     diffusion = load_model_from_config('X:\\dlas\\experiments\\train_music_diffusion_tfd_and_cheater.yml', 'generator',
                                        also_load_savepoint=False, strict_load=False,
-                                       load_path='X:\\dlas\\experiments\\train_music_diffusion_tfd14_and_cheater_g2\\models\\120000_generator_ema.pth'
+                                       load_path='X:\\dlas\\experiments\\tfd14_and_cheater.pth'
                                        ).cuda()
-    opt_eval = {'path': 'Y:\\split\\yt-music-eval',  # eval music, mostly electronica. :)
+    opt_eval = {#'path': 'Y:\\split\\yt-music-eval',  # eval music, mostly electronica. :)
                 #'path': 'E:\\music_eval',  # this is music from the training dataset, including a lot more variety.
+                'path': 'Y:\\separated\\tfd14_test',
                 'diffusion_steps': 256,
-                'conditioning_free': False, 'conditioning_free_k': 1, 'use_ddim': False, 'clip_audio': True,
+                'conditioning_free': True, 'conditioning_free_k': 1, 'use_ddim': False, 'clip_audio': True,
                 'diffusion_schedule': 'cosine', 'diffusion_type': 'from_codes_quant',
     }
 
-    env = {'rank': 0, 'base_path': 'D:\\tmp\\test_eval_music', 'step': 13, 'device': 'cuda', 'opt': {}}
+    env = {'rank': 0, 'base_path': 'D:\\tmp\\test_eval_music', 'step': 18, 'device': 'cuda', 'opt': {}}
     eval = MusicDiffusionFid(diffusion, opt_eval, env)
     fds = []
     for i in range(2):

codes/trainer/injectors/audio_injectors.py

@@ -11,7 +11,7 @@ from utils.util import opt_get, load_model_from_config, pad_or_truncate
 
 MEL_MIN = -11.512925148010254
 TACOTRON_MEL_MAX = 2.3143386840820312
-TORCH_MEL_MAX = 4.82
+TORCH_MEL_MAX = 4.82  # FYI: this STILL isn't assertive enough...
 
 def normalize_torch_mel(mel):
     return 2 * ((mel - MEL_MIN) / (TORCH_MEL_MAX - MEL_MIN)) - 1