| 198 | self.base_and_epoch_seed = seed |
| 199 | |
| 200 | def set_sequence_status(self): |
| 201 | sequence_status = dict( |
| 202 | curr = 0, |
| 203 | sample_lens = list(), |
| 204 | packed_position_ids = list(), |
| 205 | nested_attention_masks = list(), |
| 206 | split_lens = list(), |
| 207 | attn_modes = list(), |
| 208 | packed_text_ids = list(), |
| 209 | packed_text_indexes = list(), |
| 210 | packed_label_ids = list(), |
| 211 | ce_loss_indexes = list(), |
| 212 | ce_loss_weights = list(), |
| 213 | |
| 214 | |
| 215 | dino_token_seqlens = list(), |
| 216 | packed_dino_token_indexes = list(), |
| 217 | packed_depths = list(), |
| 218 | packed_extrinsics = list(), |
| 219 | packed_intrinsics = list(), |
| 220 | packed_cam_points = list(), |
| 221 | packed_world_points = list(), |
| 222 | packed_point_masks = list(), |
| 223 | packed_view_infos = list(), |
| 224 | packed_image_paths = list(), |
| 225 | packed_dino_image_tensor_list = list(), |
| 226 | packed_image_grid_thw = list(), |
| 227 | |
| 228 | packed_vit_tokens = list(), |
| 229 | packed_vit_images = list(), |
| 230 | vit_token_seqlens = list(), |
| 231 | packed_vit_token_indexes = list(), |
| 232 | img_per_seq_lens = list(), |
| 233 | ) |
| 234 | return sequence_status |
| 235 | |
| 236 | def to_tensor(self, sequence_status): |
| 237 | data = dict( |