(src_model, swiss_model)
| 51 | copy_layer_param(src.mlp.fc2, dst.mlp.dense_4h_to_h) |
| 52 | |
| 53 | def transform_weight(src_model, swiss_model): |
| 54 | copy_from_param(src_model.cls_token.data[0], swiss_model.transformer.word_embeddings.weight) |
| 55 | copy_from_param(src_model.pos_embed.data[0], swiss_model.transformer.position_embeddings.weight) |
| 56 | copy_layer_norm(src_model, swiss_model) |
| 57 | for src_l, dst_l in zip(src_model.blocks, swiss_model.transformer.layers): |
| 58 | copy_transformer_layer_wo_ln(src_l, dst_l) |
| 59 | copy_layer_param(src_model.head, swiss_model.mixins.cls.classifier) |
| 60 | copy_layer_param(src_model.patch_embed.proj, swiss_model.mixins.patch_embedding.proj) |
| 61 | |
| 62 | |
| 63 | from sat.training.model_io import save_checkpoint |
no test coverage detected