| import torch |
| import torch.nn as nn |
|
|
| from einops import rearrange |
| from diffusers import ConfigMixin, ModelMixin |
|
|
|
|
| class AudioProjModel(ModelMixin, ConfigMixin): |
| def __init__( |
| self, |
| seq_len=5, |
| seq_len_vf=12, |
| blocks=12, |
| channels=768, |
| intermediate_dim=512, |
| output_dim=768, |
| context_tokens=32, |
| norm_output_audio=True, |
| enable_compile=False, |
| ): |
| super().__init__() |
|
|
| self.seq_len = seq_len |
| self.blocks = blocks |
| self.channels = channels |
| self.input_dim = seq_len * blocks * channels |
| self.input_dim_vf = seq_len_vf * blocks * channels |
| self.intermediate_dim = intermediate_dim |
| self.context_tokens = context_tokens |
| self.output_dim = output_dim |
|
|
| |
| self.proj1 = nn.Linear(self.input_dim, intermediate_dim) |
| self.proj1_vf = nn.Linear(self.input_dim_vf, intermediate_dim) |
| self.proj2 = nn.Linear(intermediate_dim, intermediate_dim) |
| self.proj3 = nn.Linear(intermediate_dim, context_tokens * output_dim) |
| self.norm = nn.LayerNorm(output_dim) if norm_output_audio else nn.Identity() |
| self.flops = 0.0 |
| self.enable_compile = enable_compile |
|
|
| def forward(self, audio_embeds, audio_embeds_vf): |
| video_length = audio_embeds.shape[1] + audio_embeds_vf.shape[1] |
| B, _, _, S, C = audio_embeds.shape |
|
|
| |
| audio_embeds = rearrange(audio_embeds, "bz f w b c -> (bz f) w b c") |
| batch_size, window_size, blocks, channels = audio_embeds.shape |
| audio_embeds = audio_embeds.view(batch_size, window_size * blocks * channels) |
|
|
| |
| audio_embeds_vf = rearrange(audio_embeds_vf, "bz f w b c -> (bz f) w b c") |
| batch_size_vf, window_size_vf, blocks_vf, channels_vf = audio_embeds_vf.shape |
| audio_embeds_vf = audio_embeds_vf.view(batch_size_vf, window_size_vf * blocks_vf * channels_vf) |
|
|
| |
| B1, _ = audio_embeds.shape |
| audio_embeds = torch.relu(self.proj1(audio_embeds)) |
| if not self.enable_compile: |
| self.flops += B1 * self.input_dim * self.intermediate_dim * 2 |
|
|
| B1_vf, _ = audio_embeds_vf.shape |
| audio_embeds_vf = torch.relu(self.proj1_vf(audio_embeds_vf)) |
| if not self.enable_compile: |
| self.flops += B1_vf * self.input_dim_vf * self.intermediate_dim * 2 |
|
|
| audio_embeds = rearrange(audio_embeds, "(bz f) c -> bz f c", bz=B) |
| audio_embeds_vf = rearrange(audio_embeds_vf, "(bz f) c -> bz f c", bz=B) |
| audio_embeds_c = torch.concat([audio_embeds, audio_embeds_vf], dim=1) |
| batch_size_c, N_t, C_a = audio_embeds_c.shape |
| audio_embeds_c = audio_embeds_c.view(batch_size_c*N_t, C_a) |
|
|
| |
| B2, _ = audio_embeds_c.shape |
| audio_embeds_c = torch.relu(self.proj2(audio_embeds_c)) |
| if not self.enable_compile: |
| self.flops += B2 * self.intermediate_dim * self.intermediate_dim * 2 |
|
|
| |
| B3, _ = audio_embeds_c.shape |
| context_tokens = self.proj3(audio_embeds_c).reshape(batch_size_c*N_t, self.context_tokens, self.output_dim) |
| if not self.enable_compile: |
| self.flops += B3 * self.intermediate_dim * (self.context_tokens * self.output_dim) * 2 |
|
|
| |
| context_tokens = self.norm(context_tokens) |
| context_tokens = rearrange(context_tokens, "(bz f) m c -> bz f m c", f=video_length) |
|
|
| return context_tokens |
