From 96ff0008a4775e9f08d2985a92cd374ca4e001e1 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=CE=9D=CE=B1=CF=81=CE=BF=CF=85=CF=83=CE=AD=C2=B7=CE=BC?=
 =?UTF-8?q?=C2=B7=CE=B3=CE=B9=CE=BF=CF=85=CE=BC=CE=B5=CE=BC=CE=AF=C2=B7?=
 =?UTF-8?q?=CE=A7=CE=B9=CE=BD=CE=B1=CE=BA=CE=AC=CE=BD=CE=BD=CE=B1?=
 <40709280+NaruseMioShirakana@users.noreply.github.com>
Date: Sat, 6 Apr 2024 15:47:53 +0800
Subject: [PATCH] Delete GPT_SoVITS/module/embedding_onnx.py

---
 GPT_SoVITS/module/embedding_onnx.py | 64 -----------------------------
 1 file changed, 64 deletions(-)
 delete mode 100644 GPT_SoVITS/module/embedding_onnx.py

diff --git a/GPT_SoVITS/module/embedding_onnx.py b/GPT_SoVITS/module/embedding_onnx.py
deleted file mode 100644
index 2f8dcf1b..00000000
--- a/GPT_SoVITS/module/embedding_onnx.py
+++ /dev/null
@@ -1,64 +0,0 @@
-# modified from https://github.com/lifeiteng/vall-e/blob/main/valle/modules/embedding.py
-import math
-
-import torch
-from torch import nn
-
-
-class TokenEmbedding(nn.Module):
-    def __init__(
-        self,
-        embedding_dim: int,
-        vocab_size: int,
-        dropout: float = 0.0,
-    ):
-        super().__init__()
-
-        self.vocab_size = vocab_size
-        self.embedding_dim = embedding_dim
-
-        self.dropout = torch.nn.Dropout(p=dropout)
-        self.word_embeddings = nn.Embedding(self.vocab_size, self.embedding_dim)
-
-    @property
-    def weight(self) -> torch.Tensor:
-        return self.word_embeddings.weight
-
-    def embedding(self, index: int) -> torch.Tensor:
-        return self.word_embeddings.weight[index : index + 1]
-
-    def forward(self, x: torch.Tensor):
-        x = self.word_embeddings(x)
-        x = self.dropout(x)
-        return x
-
-
-class SinePositionalEmbedding(nn.Module):
-    def __init__(
-        self,
-        embedding_dim: int,
-        dropout: float = 0.0,
-        scale: bool = False,
-        alpha: bool = False,
-    ):
-        super().__init__()
-        self.embedding_dim = embedding_dim
-        self.x_scale = math.sqrt(embedding_dim) if scale else 1.0
-        self.alpha = nn.Parameter(torch.ones(1), requires_grad=alpha)
-        self.dropout = torch.nn.Dropout(p=dropout)
-        self.reverse = False
-        self.div_term = torch.exp(torch.arange(0, self.embedding_dim, 2) * -(math.log(10000.0) / self.embedding_dim))
-        self.pe = self.extend_pe(2000)
-
-    def extend_pe(self, x):
-        position = torch.cumsum(torch.ones((x,1)), dim=0)
-        scpe = (position * self.div_term).unsqueeze(0)
-        pe = torch.cat([torch.sin(scpe), torch.cos(scpe)]).permute(1, 2, 0)
-        pe = pe.contiguous().view(1, -1, self.embedding_dim)
-        return pe
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        pe = self.pe[:,:x.size(1),:]
-        output = x.unsqueeze(-1) if x.ndim == 2 else x
-        output = output * self.x_scale + self.alpha * pe
-        return self.dropout(output)