Parameter size mismatch when loading pretrained models

[qid12]

Hi, I run into errors when trying to load pretrained models.
I downloaded and processed the DB5.5 dataset, and by running get_dataloader, I could see that the 'input_edge_feats_dim' is 27 for this dataset. But then as I run:

model = create_model(args, log)
model.load_state_dict(checkpoint['state_dict'])

I got the error of 'size mismatch' for many parameters:

Here's the args I was using:
{'debug': False, 'log_every': 100000, 'random_seed': 8, 'data': 'db5', 'data_fraction': 1.0, 'split': 0, 'worker': 0, 'n_jobs': 1, 'lr': 0.0003, 'w_decay': 0.0001, 'scheduler': 'warmup', 'warmup': 1.0, 'patience': 50, 'num_epochs': 10000, 'clip': 100.0, 'bs': 10, 'graph_nodes': 'residues', 'graph_cutoff': 30.0, 'graph_max_neighbor': 10, 'graph_residue_loc_is_alphaC': False, 'pocket_cutoff': 8.0, 'translation_interval': 5.0, 'rot_model': 'kb_att', 'num_att_heads': 50, 'pocket_ot_loss_weight': 1.0, 'intersection_loss_weight': 10.0, 'intersection_sigma': 25.0, 'intersection_surface_ct': 10.0, 'dropout': 0.0, 'layer_norm': 'LN', 'layer_norm_coors': '0', 'final_h_layer_norm': '0', 'nonlin': 'lkyrelu', 'iegmn_lay_hid_dim': 64, 'iegmn_n_lays': 5, 'residue_emb_dim': 64, 'shared_layers': False, 'cross_msgs': False, 'divide_coors_dist': False, 'use_dist_in_layers': False, 'use_edge_features_in_gmn': False, 'noise_decay_rate': 0.0, 'noise_initial': 0.0, 'use_mean_node_features': False, 'skip_weight_h': 0.5, 'leakyrelu_neg_slope': 0.01, 'x_connection_init': 0.0, 'hyper_search': False, 'fine_tune': False, 'toy': False, 'continue_train_model': '', 'device': device(type='cuda', index=1), 'cache_path': './cache/db5_residues_maxneighbor_10_cutoff_30.0_pocketCut_8.0/', 'checkpoint_dir': './checkpts/EQUIDOCK__drp_0.0#Wdec_0.0001#ITS_lw_10.0#Hdim_64#Nlay_5#shrdLay_F#SURFfs_F#ln_LN#lnX_0#Hnrm_0#NattH_50#skH_0.5#xConnI_0.0#LkySl_0.01#pokOTw_1.0#divXdist_F#', 'tb_log_dir': './tb_logs/EQUIDOCK__drp_0.0#Wdec_0.0001#ITS_lw_10.0#Hdim_64#Nlay_5#shrdLay_F#SURFfs_F#ln_LN#lnX_0#Hnrm_0#NattH_50#skH_0.5#xConnI_0.0#LkySl_0.01#pokOTw_1.0#divXdist_F#', 'input_edge_feats_dim': 27, 'iegnm_lay_hid_dim': 69}

Could you please help me with this error? I wonder maybe some of the model parameters was different in the pretrained model. Please let me know what I could change to set the param size right, thank you very much!