MCPcopy Create free account
hub / github.com/VisionXLab/OF-Diff / timestep_embedding

Function timestep_embedding

ldm/modules/diffusionmodules/util.py:154–174  ·  view source on GitHub ↗

Create sinusoidal timestep embeddings. :param timesteps: a 1-D Tensor of N indices, one per batch element. These may be fractional. :param dim: the dimension of the output. :param max_period: controls the minimum frequency of the embeddings. :return: an [N

(timesteps, dim, max_period=10000, repeat_only=False)

Source from the content-addressed store, hash-verified

152
153
154def timestep_embedding(timesteps, dim, max_period=10000, repeat_only=False):
155 """
156 Create sinusoidal timestep embeddings.
157 :param timesteps: a 1-D Tensor of N indices, one per batch element.
158 These may be fractional.
159 :param dim: the dimension of the output.
160 :param max_period: controls the minimum frequency of the embeddings.
161 :return: an [N x dim] Tensor of positional embeddings.
162 """
163 if not repeat_only:
164 half = dim // 2
165 freqs = torch.exp(
166 -math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half
167 ).to(device=timesteps.device)
168 args = timesteps[:, None].float() * freqs[None]
169 embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
170 if dim % 2:
171 embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
172 else:
173 embedding = repeat(timesteps, 'b -> b d', d=dim)
174 return embedding
175
176
177def zero_module(module):

Callers 3

forwardMethod · 0.90
forwardMethod · 0.90
forwardMethod · 0.90

Calls

no outgoing calls

Tested by

no test coverage detected