(self, all_layer_hiddens, attention_mask, padding_side="left")
| 197 | self.embedding_dim = embedding_dim |
| 198 | |
| 199 | def forward(self, all_layer_hiddens, attention_mask, padding_side="left"): |
| 200 | normed = norm_and_concat_per_token_rms(all_layer_hiddens, attention_mask) |
| 201 | normed = normed.to(all_layer_hiddens.dtype) |
| 202 | v_dim = self.video_aggregate_embed.out_features |
| 203 | result = { |
| 204 | "video": self.video_aggregate_embed( |
| 205 | _rescale_norm(normed, v_dim, self.embedding_dim) |
| 206 | ) |
| 207 | } |
| 208 | if self.audio_aggregate_embed is not None: |
| 209 | a_dim = self.audio_aggregate_embed.out_features |
| 210 | result["audio"] = self.audio_aggregate_embed( |
| 211 | _rescale_norm(normed, a_dim, self.embedding_dim) |
| 212 | ) |
| 213 | return result |
| 214 | |
| 215 | |
| 216 | # --------------------------------------------------------------------------- |
nothing calls this directly
no test coverage detected