{ "adam_epsilon": 1e-08, "asl_gamma_neg": 4.0, "asl_gamma_pos": 1.0, "best_metric_type": "f1", "beta1": 0.9, "beta2": 0.99, "buffer_size": 128, "cache_dir": null, "cell_append_eos": true, "cell_embedding_complete": true, "cell_embedding_complete_seg_overlap": true, "cell_embedding_fixed_len_a_time": 9600, "cell_embedding_vector_type": "mean", "cell_llm_dirpath": "../llm//models/lucacell/lucacell-2048pos/v1.0/checkpoint-step6400000", "cell_llm_step": "6400000", "cell_llm_type": "lucacell", "cell_llm_version": "lucacell-2048pos/v1.0", "cell_matrix_add_special_token": true, "cell_matrix_dirpath": null, "cell_matrix_embedding_exists": true, "cell_max_length": 9602, "cell_max_length_a": null, "cell_max_length_b": null, "cell_no_position_embeddings": false, "cell_no_token_embeddings": true, "cell_no_token_type_embeddings": false, "cell_position_embedding_type": "RoPE", "cell_prepend_bos": true, "cell_vector_dirpath": null, "classifier_activate_func": "gelu", "classifier_size": 256, "config_path": "../config/lucasingle/lucasingle_config.json", "cross_atten": false, "dataset_name": "CRA016741_finetune_lucacellv2", "dataset_type": "cell", "delete_old": false, "device": "cuda", "do_eval": true, "do_metrics": true, "do_predict": true, "do_train": true, "dropout_prob": 0.1, "dynamic_weighted_mse_high_weight": 5.0, "dynamic_weighted_mse_low_weight": 1.0, "dynamic_weighted_mse_threshold": 0.5, "early_stop_begin_epoch": null, "early_stop_num_epoch": null, "emb_activate_func": "gelu", "embedding_input_size": 2560, "embedding_input_size_a": null, "embedding_input_size_b": null, "evaluate_during_training": true, "evaluate_steps": -1, "evaluate_strategy": "epoch", "fc_activate_func": "gelu", "finetune_layers": "20", "focal_loss_alpha": 0.7, "focal_loss_gamma": 2.0, "focal_loss_reduce": false, "fp16": false, "fp16_embedding": false, "fp16_opt_level": "O1", "fusion_type": "concat", "gene_express_vocab_size": 56, "gradient_accumulation_steps": 8, "hidden_size": 2560, "ignore_index": -100, "init_weight_filepath_for_gene_express_bin": "../llm//models/lucacell/lucacell-2048pos/v1.0/express_token_mask/20260226165643/checkpoint-step6400000/express_bin_tokens_embed.pth", "input_mode": "single", "input_type": "matrix", "intermediate_size": null, "label_filepath": "../dataset/CRA016741_finetune_lucacellv2/cell/multi_class/label.txt", "label_size": 54, "label_type": "CRA016741_finetune_lucacellv2", "learning_rate": 0.0002, "local_rank": -1, "log_dir": "../logs/CRA016741_finetune_lucacellv2/cell/multi_class/lucasingle/matrix/20260410151230", "logging_steps": 200, "loss_reduction": "mean", "loss_type": "cce", "lr_decay_rate": 0.95, "lr_update_strategy": "step", "matrix_encoder": false, "matrix_encoder_act": false, "matrix_fc_size": "null", "matrix_pooling_type": "value_attention", "max_grad_norm": 1.0, "max_steps": -1, "model_dirpath": null, "model_type": "lucasingle", "n_gpu": 1, "no_cuda": false, "non_ignore": false, "not_cell_append_eos": false, "not_cell_prepend_bos": false, "not_frozen_gene_express_bin_embedding": true, "not_matrix_encoder_shared": false, "not_save_emb_to_disk": false, "not_seq_append_eos": false, "not_seq_prepend_bos": false, "num_attention_heads": 8, "num_hidden_layers": 1, "num_labels": 54, "num_train_epochs": 1000, "output_dir": "../models/CRA016741_finetune_lucacellv2/cell/multi_class/lucasingle/matrix/20260410151230", "output_mode": "multi_class", "per_gpu_eval_batch_size": 1, "per_gpu_train_batch_size": 1, "pos_weight": 1.0, "position_embedding_type": "RoPE", "save_all": true, "save_steps": -1, "seed": 1221, "self_atten": false, "seq_append_eos": true, "seq_buffer_size": 1280000, "seq_embedding_complete": true, "seq_embedding_complete_seg_overlap": true, "seq_embedding_fixed_len_a_time": 10240, "seq_embedding_vector_type": "mean", "seq_llm_dirpath": "../llm//models/lucaone/lucaone/checkpoint-step60000000", "seq_llm_step": "60000000", "seq_llm_type": "lucaone", "seq_llm_version": "lucaone", "seq_matrix_add_special_token": true, "seq_matrix_dirpath": null, "seq_matrix_embedding_exists": false, "seq_max_length": 10242, "seq_meta_fasta": "../../data/microbe/PRJCA017256/CRA016741/sampled_renamed_seqs.all.fa", "seq_no_position_embeddings": false, "seq_no_token_embeddings": true, "seq_no_token_type_embeddings": false, "seq_position_embedding_type": "RoPE", "seq_prepend_bos": true, "seq_vector_dirpath": "../../embedding/lucacellv2/lucacell-2048pos/v1.0/CRA016741_finetune/seq_embeddings1/#../../embedding/lucacellv2/lucacell-2048pos/v1.0/CRA016741_finetune/seq_embeddings/", "seq_vector_embedding_exists": false, "sigmoid": false, "task_level_type": "cell_level", "task_type": "multi_class", "tb_log_dir": "../tb-logs/CRA016741_finetune_lucacellv2/cell/multi_class/lucasingle/matrix/20260410151230", "test_data_dir": "../dataset/CRA016741_finetune_lucacellv2/cell/multi_class/test/", "time_str": "20260410151234", "train_data_dir": "../dataset/CRA016741_finetune_lucacellv2/cell/multi_class/train/", "trunc_type": "right", "use_embed_layer_norm": false, "use_rotary_position_embeddings_for_cross": false, "val_data_dir": "../dataset/CRA016741_finetune_lucacellv2/cell/multi_class/val/", "vector_fc_size": "null", "warmup_steps": 1000, "weight": [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "weight_decay": 0.01, "worker_num": 0 } ################################################## n_gpu: 1 ################################################## Inputs: Input Name List: cell,cell_embedding_matrix ################################################## Encoder Config: {'input_type': 'matrix', 'trunc_type': 'right', 'seq_llm_dirpath': '../llm//models/lucaone/lucaone/checkpoint-step60000000', 'seq_llm_type': 'lucaone', 'seq_llm_version': 'lucaone', 'seq_llm_step': '60000000', 'seq_tokenizer': , 'seq_max_length': 10242, 'seq_prepend_bos': True, 'seq_append_eos': True, 'seq_matrix_add_special_token': True, 'seq_vector_dirpath': '../../embedding/lucacellv2/lucacell-2048pos/v1.0/CRA016741_finetune/seq_embeddings1/#../../embedding/lucacellv2/lucacell-2048pos/v1.0/CRA016741_finetune/seq_embeddings/', 'seq_matrix_dirpath': None, 'seq_embedding_vector_type': 'mean', 'seq_embedding_complete': True, 'seq_embedding_complete_seg_overlap': True, 'seq_embedding_fixed_len_a_time': 10240, 'seq_vector_embedding_exists': False, 'seq_matrix_embedding_exists': False, 'seq_meta_fasta': '../../data/microbe/PRJCA017256/CRA016741/sampled_renamed_seqs.all.fa', 'cell_llm_dirpath': '../llm//models/lucacell/lucacell-2048pos/v1.0/checkpoint-step6400000', 'cell_llm_type': 'lucacell', 'cell_llm_version': 'lucacell-2048pos/v1.0', 'cell_llm_step': '6400000', 'cell_tokenizer': , 'cell_max_length': 9602, 'cell_prepend_bos': True, 'cell_append_eos': True, 'cell_matrix_add_special_token': True, 'cell_vector_dirpath': None, 'cell_matrix_dirpath': None, 'cell_embedding_vector_type': 'mean', 'cell_embedding_complete': True, 'cell_embedding_complete_seg_overlap': True, 'cell_embedding_fixed_len_a_time': 9600, 'cell_matrix_embedding_exists': True, 'gpu_id': 0, 'lucacell_finetune': False, 'not_frozen_gene_express_bin_embedding': True, 'buffer_size': 128, 'seq_buffer_size': 1280000} ################################################## Model Config: LucaConfig { "alphabet": "gene_prot", "attention_probs_dropout_prob": 0.1, "cell_max_length": 9602, "cell_no_position_embeddings": false, "cell_no_token_embeddings": true, "cell_no_token_type_embeddings": false, "cell_position_embedding_type": "RoPE", "classifier_activate_func": "gelu", "classifier_dropout_prob": 0.1, "classifier_size": 256, "cls_token_id": 2, "cross_atten": false, "directionality": "bidi", "emb_activate_func": "gelu", "embedding_input_size": 2560, "fc_activate_func": "gelu", "gene_express_vocab_size": 56, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "hidden_size": 2560, "id2label": {}, "ignore_index": -100, "initializer_range": 0.02, "intermediate_size": 10240, "kernel_size": 7, "label2id": {}, "layer_norm_eps": 1e-06, "layer_norm_name": "norm", "layer_norm_type": "post", "loss_reduction": "mean", "matrix_fc_size": null, "matrix_pooling_type": "value_attention", "max_position_embeddings": 2048, "no_position_embeddings": true, "no_token_embeddings": false, "no_token_type_embeddings": true, "num_attention_heads": 8, "num_hidden_layers": 1, "pad_token_id": 0, "pos_weight": 1.0, "position_embedding_type": "RoPE", "self_atten": false, "sep_token_id": 3, "seq_fc_size": null, "seq_max_length": 10242, "seq_no_position_embeddings": false, "seq_no_token_embeddings": true, "seq_no_token_type_embeddings": false, "seq_position_embedding_type": "RoPE", "token_dropout": null, "transformers_version": "4.57.0", "type_vocab_size": 2, "use_embed_layer_norm": false, "use_rotary_position_embeddings": true, "use_rotary_position_embeddings_for_cross": false, "vector_fc_size": null, "vocab_size": null, "weight": [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ] } ################################################## Mode Architecture: LucaSingle( (gene_express_embedding): LucaEmbeddings( (word_embeddings): Embedding(56, 2560, padding_idx=0) (dropout): Dropout(p=0.1, inplace=False) ) (matrix_encoder): LucaTransformer( (embeddings): LucaEmbeddings( (word_embeddings): Embedding(56, 2560, padding_idx=0) (dropout): Dropout(p=0.1, inplace=False) ) (embed_layer_norm): LayerNorm((2560,), eps=1e-05, elementwise_affine=True) (encoder): ModuleList( (0): LucaTransformerLayer( (self_attn): LucaMultiHeadAttention( (k_proj): Linear(in_features=2560, out_features=2560, bias=True) (v_proj): Linear(in_features=2560, out_features=2560, bias=True) (q_proj): Linear(in_features=2560, out_features=2560, bias=True) (out_proj): Linear(in_features=2560, out_features=2560, bias=True) (rot_emb): RotaryEmbedding() ) (self_attn_layer_norm): LayerNorm((2560,), eps=1e-05, elementwise_affine=True) (fc1): Linear(in_features=2560, out_features=10240, bias=True) (fc2): Linear(in_features=10240, out_features=2560, bias=True) (final_layer_norm): LayerNorm((2560,), eps=1e-05, elementwise_affine=True) ) ) ) (matrix_pooler): GlobalMaskValueAttentionPooling1D(2560 -> 2560, bias=(False, False)) (dropout): Dropout(p=0.1, inplace=False) (hidden_layer): Linear(in_features=2560, out_features=256, bias=True) (hidden_act): GELU(approximate='none') (classifier): Linear(in_features=256, out_features=54, bias=True) (output): Softmax(dim=-1) (loss_fct): MaskedCrossEntropyLoss( (criterion): CrossEntropyLoss() ) ) ################################################## Model parameters: 99155254 ################################################## {"total_num": "94.560000M", "total_size": "378.250000MB", "param_sum": "94.560000M", "param_size": "378.250000MB", "buffer_sum": "0.000000M", "buffer_size": "0.000000MB", "trainable_num": "94.561819M", "trainable_size": "378.247276MB"} ##################################################