MCPcopy Create free account
hub / github.com/FireRedTeam/FireRedTTS2 / forward_chunk

Method forward_chunk

fireredtts2/codec/decoder.py:150–171  ·  view source on GitHub ↗

Args: x: shape (b, c, t) cache: shape (b, c_in+c_out, t=2)

(self, x: torch.Tensor, cache: torch.Tensor = None)

Source from the content-addressed store, hash-verified

148 return x + h
149
150 def forward_chunk(self, x: torch.Tensor, cache: torch.Tensor = None):
151 """
152 Args:
153 x: shape (b, c, t)
154 cache: shape (b, c_in+c_out, t=2)
155 """
156 cache1, cache2 = (
157 (None, None)
158 if cache is None
159 else cache.split((self.in_channels, self.out_channels), dim=1)
160 )
161 h = x
162 # block1
163 h = self.block1[:4](h)
164 h, new_cache1 = self.block1[4].forward_chunk(h, cache1)
165 # block2
166 h = self.block2[:5](h)
167 h, new_cache2 = self.block2[5].forward_chunk(h, cache2)
168 if self.in_channels != self.out_channels:
169 x = self.nin_shortcut(x)
170 new_cache = torch.cat([new_cache1, new_cache2], dim=1)
171 return x + h, new_cache
172
173
174# Nonstreaming Vocos backbone based on Transformer layers

Callers

nothing calls this directly

Calls 1

forward_chunkMethod · 0.45

Tested by

no test coverage detected