(self, add_positional_conv=None)
| 237 | |
| 238 | class SVDUNet(torch.nn.Module): |
| 239 | def __init__(self, add_positional_conv=None): |
| 240 | super().__init__() |
| 241 | self.time_proj = Timesteps(320) |
| 242 | self.time_embedding = torch.nn.Sequential( |
| 243 | torch.nn.Linear(320, 1280), |
| 244 | torch.nn.SiLU(), |
| 245 | torch.nn.Linear(1280, 1280) |
| 246 | ) |
| 247 | self.add_time_proj = Timesteps(256) |
| 248 | self.add_time_embedding = torch.nn.Sequential( |
| 249 | torch.nn.Linear(768, 1280), |
| 250 | torch.nn.SiLU(), |
| 251 | torch.nn.Linear(1280, 1280) |
| 252 | ) |
| 253 | self.conv_in = torch.nn.Conv2d(8, 320, kernel_size=3, padding=1) |
| 254 | |
| 255 | self.blocks = torch.nn.ModuleList([ |
| 256 | # CrossAttnDownBlockSpatioTemporal |
| 257 | ResnetBlock(320, 320, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(320, 320, 1280, eps=1e-6), PopMixBlock(), PushBlock(), |
| 258 | AttentionBlock(5, 64, 320, 1, 1024, need_proj_out=False), PushBlock(), TemporalAttentionBlock(5, 64, 320, 1024, add_positional_conv), PopMixBlock(320), PushBlock(), |
| 259 | ResnetBlock(320, 320, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(320, 320, 1280, eps=1e-6), PopMixBlock(), PushBlock(), |
| 260 | AttentionBlock(5, 64, 320, 1, 1024, need_proj_out=False), PushBlock(), TemporalAttentionBlock(5, 64, 320, 1024, add_positional_conv), PopMixBlock(320), PushBlock(), |
| 261 | DownSampler(320), PushBlock(), |
| 262 | # CrossAttnDownBlockSpatioTemporal |
| 263 | ResnetBlock(320, 640, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(640, 640, 1280, eps=1e-6), PopMixBlock(), PushBlock(), |
| 264 | AttentionBlock(10, 64, 640, 1, 1024, need_proj_out=False), PushBlock(), TemporalAttentionBlock(10, 64, 640, 1024, add_positional_conv), PopMixBlock(640), PushBlock(), |
| 265 | ResnetBlock(640, 640, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(640, 640, 1280, eps=1e-6), PopMixBlock(), PushBlock(), |
| 266 | AttentionBlock(10, 64, 640, 1, 1024, need_proj_out=False), PushBlock(), TemporalAttentionBlock(10, 64, 640, 1024, add_positional_conv), PopMixBlock(640), PushBlock(), |
| 267 | DownSampler(640), PushBlock(), |
| 268 | # CrossAttnDownBlockSpatioTemporal |
| 269 | ResnetBlock(640, 1280, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(1280, 1280, 1280, eps=1e-6), PopMixBlock(), PushBlock(), |
| 270 | AttentionBlock(20, 64, 1280, 1, 1024, need_proj_out=False), PushBlock(), TemporalAttentionBlock(20, 64, 1280, 1024, add_positional_conv), PopMixBlock(1280), PushBlock(), |
| 271 | ResnetBlock(1280, 1280, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(1280, 1280, 1280, eps=1e-6), PopMixBlock(), PushBlock(), |
| 272 | AttentionBlock(20, 64, 1280, 1, 1024, need_proj_out=False), PushBlock(), TemporalAttentionBlock(20, 64, 1280, 1024, add_positional_conv), PopMixBlock(1280), PushBlock(), |
| 273 | DownSampler(1280), PushBlock(), |
| 274 | # DownBlockSpatioTemporal |
| 275 | ResnetBlock(1280, 1280, 1280, eps=1e-5), PushBlock(), TemporalResnetBlock(1280, 1280, 1280, eps=1e-5), PopMixBlock(), PushBlock(), |
| 276 | ResnetBlock(1280, 1280, 1280, eps=1e-5), PushBlock(), TemporalResnetBlock(1280, 1280, 1280, eps=1e-5), PopMixBlock(), PushBlock(), |
| 277 | # UNetMidBlockSpatioTemporal |
| 278 | ResnetBlock(1280, 1280, 1280, eps=1e-5), PushBlock(), TemporalResnetBlock(1280, 1280, 1280, eps=1e-5), PopMixBlock(), PushBlock(), |
| 279 | AttentionBlock(20, 64, 1280, 1, 1024, need_proj_out=False), PushBlock(), TemporalAttentionBlock(20, 64, 1280, 1024, add_positional_conv), PopMixBlock(1280), |
| 280 | ResnetBlock(1280, 1280, 1280, eps=1e-5), PushBlock(), TemporalResnetBlock(1280, 1280, 1280, eps=1e-5), PopMixBlock(), |
| 281 | # UpBlockSpatioTemporal |
| 282 | PopBlock(), ResnetBlock(2560, 1280, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(1280, 1280, 1280, eps=1e-5), PopMixBlock(), |
| 283 | PopBlock(), ResnetBlock(2560, 1280, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(1280, 1280, 1280, eps=1e-5), PopMixBlock(), |
| 284 | PopBlock(), ResnetBlock(2560, 1280, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(1280, 1280, 1280, eps=1e-5), PopMixBlock(), |
| 285 | UpSampler(1280), |
| 286 | # CrossAttnUpBlockSpatioTemporal |
| 287 | PopBlock(), ResnetBlock(2560, 1280, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(1280, 1280, 1280, eps=1e-6), PopMixBlock(), PushBlock(), |
| 288 | AttentionBlock(20, 64, 1280, 1, 1024, need_proj_out=False), PushBlock(), TemporalAttentionBlock(20, 64, 1280, 1024, add_positional_conv), PopMixBlock(1280), |
| 289 | PopBlock(), ResnetBlock(2560, 1280, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(1280, 1280, 1280, eps=1e-6), PopMixBlock(), PushBlock(), |
| 290 | AttentionBlock(20, 64, 1280, 1, 1024, need_proj_out=False), PushBlock(), TemporalAttentionBlock(20, 64, 1280, 1024, add_positional_conv), PopMixBlock(1280), |
| 291 | PopBlock(), ResnetBlock(1920, 1280, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(1280, 1280, 1280, eps=1e-6), PopMixBlock(), PushBlock(), |
| 292 | AttentionBlock(20, 64, 1280, 1, 1024, need_proj_out=False), PushBlock(), TemporalAttentionBlock(20, 64, 1280, 1024, add_positional_conv), PopMixBlock(1280), |
| 293 | UpSampler(1280), |
| 294 | # CrossAttnUpBlockSpatioTemporal |
| 295 | PopBlock(), ResnetBlock(1920, 640, 1280, eps=1e-6), PushBlock(), TemporalResnetBlock(640, 640, 1280, eps=1e-6), PopMixBlock(), PushBlock(), |
| 296 | AttentionBlock(10, 64, 640, 1, 1024, need_proj_out=False), PushBlock(), TemporalAttentionBlock(10, 64, 640, 1024, add_positional_conv), PopMixBlock(640), |
no test coverage detected