Disable autcast for criterions

TTS/tts/models/fast_pitch.py

@@ -1,9 +1,10 @@
 from dataclasses import dataclass, field
+from typing import Tuple
 
 import torch
-import torch.nn.functional as F
 from coqpit import Coqpit
 from torch import nn
+from torch.cuda.amp.autocast_mode import autocast
 
 from TTS.tts.layers.feed_forward.decoder import Decoder
 from TTS.tts.layers.feed_forward.encoder import Encoder
@@ -12,7 +13,6 @@ from TTS.tts.layers.glow_tts.duration_predictor import DurationPredictor
 from TTS.tts.layers.glow_tts.monotonic_align import generate_path, maximum_path
 from TTS.tts.models.base_tts import BaseTTS
 from TTS.tts.utils.data import sequence_mask
-from TTS.tts.utils.measures import alignment_diagonal_score
 from TTS.tts.utils.visual import plot_alignment, plot_spectrogram
 from TTS.utils.audio import AudioProcessor
 
@@ -30,7 +30,7 @@ class AlignmentEncoder(torch.nn.Module):
         self.softmax = torch.nn.Softmax(dim=3)
         self.log_softmax = torch.nn.LogSoftmax(dim=3)
 
-        self.key_proj = nn.Sequential(
+        self.key_layer = nn.Sequential(
             nn.Conv1d(
                 in_key_channels,
                 in_key_channels * 2,
@@ -42,7 +42,7 @@ class AlignmentEncoder(torch.nn.Module):
             nn.Conv1d(in_key_channels * 2, attn_channels, kernel_size=1, padding=0, bias=True),
         )
 
-        self.query_proj = nn.Sequential(
+        self.query_layer = nn.Sequential(
             nn.Conv1d(
                 in_query_channels,
                 in_query_channels * 2,
@@ -58,33 +58,26 @@ class AlignmentEncoder(torch.nn.Module):
 
     def forward(
         self, queries: torch.tensor, keys: torch.tensor, mask: torch.tensor = None, attn_prior: torch.tensor = None
-    ):
+    ) -> Tuple[torch.tensor, torch.tensor]:
         """Forward pass of the aligner encoder.
         Shapes:
-            - queries: :math:`(B, C, T_de)`
+            - queries: :math:`[B, C, T_de]`
-            - keys: :math:`(B, C_emb, T_en)`
+            - keys: :math:`[B, C_emb, T_en]`
-            - mask: :math:`(B, T_de)`
+            - mask: :math:`[B, T_de]`
         Output:
-            attn (torch.tensor): B x 1 x T1 x T2 attention mask. Final dim T2 should sum to 1.
+            attn (torch.tensor): :math:`[B, 1, T_en, T_de]` soft attention mask.
-            attn_logprob (torch.tensor): B x 1 x T1 x T2 log-prob attention mask.
+            attn_logp (torch.tensor): :math:`[ßB, 1, T_en , T_de]` log probabilities.
         """
-        keys_enc = self.key_proj(keys)  # B x n_attn_dims x T2
+        key_out = self.key_layer(keys)
-        queries_enc = self.query_proj(queries)
+        query_out = self.query_layer(queries)
-
+        attn_factor = (query_out[:, :, :, None] - key_out[:, :, None]) ** 2
-        # Simplistic Gaussian Isotopic Attention
+        attn_factor = -self.temperature * attn_factor.sum(1, keepdim=True)
-        attn = (queries_enc[:, :, :, None] - keys_enc[:, :, None]) ** 2  # B x n_attn_dims x T1 x T2
-        attn = -self.temperature * attn.sum(1, keepdim=True)
-
         if attn_prior is not None:
-            attn = self.log_softmax(attn) + torch.log(attn_prior[:, None] + 1e-8)
+            attn_logp = self.log_softmax(attn_factor) + torch.log(attn_prior[:, None] + 1e-8)
-
-        attn_logprob = attn.clone()
-
         if mask is not None:
-            attn.data.masked_fill_(~mask.bool().unsqueeze(2), -float("inf"))
+            attn_logp.data.masked_fill_(~mask.bool().unsqueeze(2), -float("inf"))
-
+        attn = self.softmax(attn_logp)
-        attn = self.softmax(attn)  # softmax along T2
+        return attn, attn_logp
-        return attn, attn_logprob
 
 
 @dataclass
@@ -414,6 +407,7 @@ class FastPitch(BaseTTS):
         if self.use_aligner:
             durations = outputs["o_alignment_dur"]
 
+        with autocast(enabled=False):  # use float32 for the criterion
             # compute loss
             loss_dict = criterion(
                 outputs["model_outputs"],