| { |
| "_attn_implementation_autoset": true, |
| "model_config": { |
| "attention_map": "softmax", |
| "attention_method": "linear", |
| "dim_embed": 256, |
| "dim_feedforward": 1024, |
| "dim_ffn_u_model": 1024, |
| "dim_hidden_u_model": 256, |
| "dim_max_trajectory": 3, |
| "dropout": 0.1, |
| "num_context_encoder_layers": 2, |
| "num_heads": 8, |
| "num_res_layer_u_model": 6, |
| "num_res_layers_functional_decoder": 8, |
| "use_bias_for_projection": true, |
| "use_bias_in_attention": true, |
| "use_query_residual_in_attention": true |
| }, |
| "train_config": { |
| "corruption_model_type": "odeformer", |
| "loss_filter_nans": true, |
| "loss_type": "l1", |
| "max_sigma_trajectory_noise": 0.06, |
| "max_subsampling_ration": 0.5, |
| "train_type": "vector_field", |
| "train_with_normalized_head": true |
| }, |
| "transformers_version": "4.46.0" |
| } |
|
|
