2025-06-03 20:22:05,457 - INFO - ๐Ÿ› ๏ธ Command-line Arguments: - [multilabel_classify.py:323:print_args] 2025-06-03 20:22:05,457 - INFO - ๐Ÿ”น output_dir: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b ๐Ÿ”น source_url: XURLs.MIMIC4_DEMO ๐Ÿ”น data: mimic4_icd10_full ๐Ÿ”น logfile: classification_log ๐Ÿ”น base_dir: ../tmp/MIMIC4_DEMO ๐Ÿ”น hub_model_id: deb101/mistral-7b-instruct-v0.3-mimic4-adapt ๐Ÿ”น model_name: mistralai/Mistral-7B-Instruct-v0.3 ๐Ÿ”น max_length: 512 ๐Ÿ”น do_fresh_training: True ๐Ÿ”น load_from_checkpoint: False ๐Ÿ”น task: multilabel-classify ๐Ÿ”น num_train_epochs: 4 ๐Ÿ”น metric_for_best_model: precision_at_15 ๐Ÿ”น learning_rate: 0.0001 ๐Ÿ”น final_lr_scheduling: 1e-06 ๐Ÿ”น warmup_steps: 500 ๐Ÿ”น source: /home/ubuntu/.xcube/data/mimic4_demo ๐Ÿ”น logfile_path: ../tmp/MIMIC4_DEMO/logs/classification_log_2025-06-03_20-22-05.log - [multilabel_classify.py:324:print_args] 2025-06-03 20:22:05,457 - INFO - โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž–โž– - [multilabel_classify.py:325:print_args] 2025-06-03 20:22:05,458 - INFO - +----------------------------------------------------+ - [multilabel_classify.py:133:log_with_box] 2025-06-03 20:22:05,458 - INFO - | ๐Ÿค–๐Ÿท๏ธ ****Multi-Label Classification Started**** ๐Ÿท๏ธ๐Ÿค– | - [multilabel_classify.py:133:log_with_box] 2025-06-03 20:22:05,458 - INFO - +----------------------------------------------------+ - [multilabel_classify.py:133:log_with_box] 2025-06-03 20:22:05,458 - INFO - ================================================================================ - [multilabel_classify.py:96:log_section] 2025-06-03 20:22:05,458 - INFO - = ๐Ÿ“Œ INITIALIZING TRAINING ENVIRONMENT = - [multilabel_classify.py:97:log_section] 2025-06-03 20:22:05,458 - INFO - ================================================================================ - [multilabel_classify.py:100:log_section] 2025-06-03 20:22:05,458 - INFO - Hub Model ID for this multi-label classification task: deb101/mistral-7b-instruct-v0.3-mimic4-adapt-multilabel-classify - [multilabel_classify.py:3227:main] 2025-06-03 20:22:05,504 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:96:log_section] 2025-06-03 20:22:05,505 - INFO - + โœจ LOADING DATASETS + - [multilabel_classify.py:97:log_section] 2025-06-03 20:22:05,505 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:100:log_section] 2025-06-03 20:22:05,505 - INFO - ๐Ÿ“Š Loading main datasets.... - [multilabel_classify.py:3245:main] 2025-06-03 20:22:13,776 - INFO - ๐Ÿ” Total unique labels in dataset: 7942 - [multilabel_classify.py:3086:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,789 - INFO - ๐Ÿงช Attempt 1: Sampled 122 rows covering 863 labels. - [multilabel_classify.py:3100:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,798 - INFO - ๐Ÿงช Attempt 2: Sampled 122 rows covering 816 labels. - [multilabel_classify.py:3100:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,807 - INFO - ๐Ÿงช Attempt 3: Sampled 122 rows covering 885 labels. - [multilabel_classify.py:3100:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,816 - INFO - ๐Ÿงช Attempt 4: Sampled 122 rows covering 828 labels. - [multilabel_classify.py:3100:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,825 - INFO - ๐Ÿงช Attempt 5: Sampled 122 rows covering 879 labels. - [multilabel_classify.py:3100:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,834 - INFO - ๐Ÿงช Attempt 6: Sampled 122 rows covering 852 labels. - [multilabel_classify.py:3100:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,842 - INFO - ๐Ÿงช Attempt 7: Sampled 122 rows covering 838 labels. - [multilabel_classify.py:3100:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,850 - INFO - ๐Ÿงช Attempt 8: Sampled 122 rows covering 851 labels. - [multilabel_classify.py:3100:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,859 - INFO - ๐Ÿงช Attempt 9: Sampled 122 rows covering 825 labels. - [multilabel_classify.py:3100:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,867 - INFO - ๐Ÿงช Attempt 10: Sampled 122 rows covering 833 labels. - [multilabel_classify.py:3100:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,868 - INFO - โš ๏ธ Skipping label coverage fix. 7109 labels are missing. - [multilabel_classify.py:3118:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,868 - INFO - โœ… Final row count: 122 (Valid: 20, Not-valid: 102) - [multilabel_classify.py:3121:sample_df_with_full_label_coverage] 2025-06-03 20:22:13,886 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:96:log_section] 2025-06-03 20:22:13,887 - INFO - + โœจ STARTING FRESH TRAINING + - [multilabel_classify.py:97:log_section] 2025-06-03 20:22:13,887 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:100:log_section] 2025-06-03 20:22:13,887 - INFO - ๐Ÿ”„ Starting fresh training (either forced or model not found)... - [multilabel_classify.py:3266:main] 2025-06-03 20:22:13,902 - WARNING - Note: Environment variable`HF_TOKEN` is set and is the current active token independently from the token you've just configured. - [_login.py:415:_login] 2025-06-03 20:22:13,902 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:96:log_section] 2025-06-03 20:22:13,902 - INFO - + โœจ LOADING BASE MODEL + - [multilabel_classify.py:97:log_section] 2025-06-03 20:22:13,902 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:100:log_section] 2025-06-03 20:22:13,902 - INFO - ๐Ÿ“ฅ Loading pretrained model and tokenizer... - [multilabel_classify.py:3296:main] 2025-06-03 20:22:13,902 - INFO - ๐Ÿš€ Starting model and tokenizer loading process... - [multilabel_classify.py:1198:load_base_model_and_tokenizer] 2025-06-03 20:22:13,903 - INFO - ๐Ÿ“Š Quantization config: BitsAndBytesConfig { "_load_in_4bit": true, "_load_in_8bit": false, "bnb_4bit_compute_dtype": "bfloat16", "bnb_4bit_quant_storage": "uint8", "bnb_4bit_quant_type": "nf4", "bnb_4bit_use_double_quant": true, "llm_int8_enable_fp32_cpu_offload": false, "llm_int8_has_fp16_weight": false, "llm_int8_skip_modules": null, "llm_int8_threshold": 6.0, "load_in_4bit": true, "load_in_8bit": false, "quant_method": "bitsandbytes" } - [multilabel_classify.py:1207:load_base_model_and_tokenizer] 2025-06-03 20:22:13,904 - INFO - ๐Ÿ”ค Loading tokenizer for model: deb101/mistral-7b-instruct-v0.3-mimic4-adapt... - [multilabel_classify.py:1211:load_base_model_and_tokenizer] 2025-06-03 20:22:14,413 - INFO - ๐Ÿ” Checking if deb101/mistral-7b-instruct-v0.3-mimic4-adapt is a PEFT model... - [multilabel_classify.py:1223:load_base_model_and_tokenizer] 2025-06-03 20:22:14,438 - INFO - โœ… Detected PEFT model. Base model: mistralai/Mistral-7B-Instruct-v0.3 - [multilabel_classify.py:1227:load_base_model_and_tokenizer] 2025-06-03 20:22:14,438 - INFO - ๐Ÿ” Loading model configuration for mistralai/Mistral-7B-Instruct-v0.3... - [multilabel_classify.py:1237:load_base_model_and_tokenizer] 2025-06-03 20:22:14,461 - INFO - Model type: mistral, Architectures: ['MistralForCausalLM'] - [multilabel_classify.py:1243:load_base_model_and_tokenizer] 2025-06-03 20:22:14,461 - INFO - ๐Ÿง  Loading base model: mistralai/Mistral-7B-Instruct-v0.3... - [multilabel_classify.py:1306:load_base_model_and_tokenizer] 2025-06-03 20:22:15,021 - INFO - We will use 90% of the memory on device 0 for storing the model, and 10% for the buffer to avoid OOM. You can set `max_memory` in to a higher value to use more memory (at your own risk). - [modeling.py:991:get_balanced_memory] 2025-06-03 20:22:20,344 - INFO - ๐Ÿงฉ Loading PEFT adapters for deb101/mistral-7b-instruct-v0.3-mimic4-adapt... - [multilabel_classify.py:1321:load_base_model_and_tokenizer] 2025-06-03 20:22:20,612 - INFO - ๐Ÿ”ง Before enabling PEFT adapters for training - [multilabel_classify.py:1323:load_base_model_and_tokenizer] 2025-06-03 20:22:20,617 - INFO - ๐Ÿ”ง After Enabling PEFT adapters for training - [multilabel_classify.py:1330:load_base_model_and_tokenizer] 2025-06-03 20:22:20,620 - INFO - โœ… Model and tokenizer successfully loaded! - [multilabel_classify.py:1371:load_base_model_and_tokenizer] 2025-06-03 20:22:20,620 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:96:log_section] 2025-06-03 20:22:20,620 - INFO - + โœจ DATA PREPROCESSING + - [multilabel_classify.py:97:log_section] 2025-06-03 20:22:20,620 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:100:log_section] 2025-06-03 20:22:20,620 - INFO - ๐Ÿ”„ Loading and preprocessing training data... - [multilabel_classify.py:3304:main] 2025-06-03 20:22:20,624 - INFO - Total number of labels: 833 - [multilabel_classify.py:999:preprocess_data] 2025-06-03 20:22:20,624 - INFO - Rare labels (freq < 50): 832 - [multilabel_classify.py:1000:preprocess_data] 2025-06-03 20:22:20,624 - INFO - Not rare labels (freq >= 50): 1 - [multilabel_classify.py:1001:preprocess_data] 2025-06-03 20:22:20,624 - INFO - Label partitions and classes saved to ../tmp/MIMIC4_DEMO/labels_partition.json - [multilabel_classify.py:1002:preprocess_data] 2025-06-03 20:22:21,956 - INFO - The size of training set: 567 - [multilabel_classify.py:1098:preprocess_data] 2025-06-03 20:22:21,956 - INFO - The size of Evaluation set: 136 - [multilabel_classify.py:1099:preprocess_data] 2025-06-03 20:22:21,961 - INFO - Number of unique ICD-10 codes: 833 - [multilabel_classify.py:3310:main] 2025-06-03 20:22:21,961 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:96:log_section] 2025-06-03 20:22:21,961 - INFO - + โœจ MODEL INITIALIZATION + - [multilabel_classify.py:97:log_section] 2025-06-03 20:22:21,961 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:100:log_section] 2025-06-03 20:22:21,961 - INFO - ๐Ÿง  Initializing custom L2R model for outputting per-token relevance scores per ICD-10 codes. - [multilabel_classify.py:3313:main] 2025-06-03 20:22:21,962 - INFO - Will now start to create Multilabel-Classification Model from the base model - [multilabel_classify.py:516:__init__] 2025-06-03 20:22:21,965 - INFO - Trainable params: 6815744 / 3765178368 (0.18%) - [multilabel_classify.py:568:compute_trainable_params] 2025-06-03 20:22:22,304 - INFO - Creating the Multi-Label Classification Model from base model mistralai/Mistral-7B-Instruct-v0.3 completed!!! - [multilabel_classify.py:558:__init__] 2025-06-03 20:22:22,307 - INFO - Trainable params: 84177025 / 3842539649 (2.19%) - [multilabel_classify.py:568:compute_trainable_params] 2025-06-03 20:22:22,307 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:96:log_section] 2025-06-03 20:22:22,307 - INFO - + โœจ TRAINING PREPARATION + - [multilabel_classify.py:97:log_section] 2025-06-03 20:22:22,307 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:100:log_section] 2025-06-03 20:22:22,307 - INFO - โš™๏ธ Preparing training components and optimizers... - [multilabel_classify.py:3320:main] 2025-06-03 20:22:22,367 - INFO - ๐Ÿ–ฅ๏ธ Device: NVIDIA GH200 480GB - [multilabel_classify.py:845:log_training_configuration] 2025-06-03 20:22:22,367 - INFO - ๐Ÿ”‹ CUDA Available: True - [multilabel_classify.py:848:log_training_configuration] 2025-06-03 20:22:22,367 - INFO - ๐Ÿ’พ CUDA Device Count: 1 - [multilabel_classify.py:849:log_training_configuration] 2025-06-03 20:22:22,368 - INFO - ๐Ÿ“‹ Training Configuration ๐Ÿ“‹ +----------+-----------------------------+------------------------------------------------------------------+ | ๐ŸŒŸ Emoji | ๐Ÿท๏ธ Parameter | ๐Ÿ“Š Value | +----------+-----------------------------+------------------------------------------------------------------+ | ๐Ÿ“ | Output Directory | ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b | | ๐Ÿ” | Training Epochs | 4 | | ๐Ÿ‹๏ธ | Train Batch Size | 8 | | ๐Ÿ” | Eval Batch Size | 8 | | ๐Ÿ“Š | Gradient Accumulation Steps | 4 | | ๐Ÿš€ | Learning Rate | 0.0001 | | ๐ŸŒ… | Warmup Steps | 500 | | ๐Ÿ’พ | Save Strategy | epoch | | ๐Ÿ’พ | Save Total Limit | 10 | | ๐Ÿ“Š | Evaluation Strategy | epoch | | ๐ŸŽฏ | Best Model Metric | precision_at_15 | | ๐Ÿ“ | Logging Strategy | steps (every 10 steps) | | ๐ŸŒ | Push to Hub | True | | ๐ŸŒ | Hub Model ID | deb101/mistral-7b-instruct-v0.3-mimic4-adapt-multilabel-classify | | ๐Ÿ”ข | Steps per Epoch | 17 | | ๐Ÿ”ข | Total Training Steps | 68 | | ๐Ÿ”ข | Evaluation Steps | 17 | | ๐Ÿ“Š | Training Dataset Size | 567 samples ๐Ÿ‹๏ธ | | ๐Ÿ“Š | Evaluation Dataset Size | 136 samples ๐Ÿ” | +----------+-----------------------------+------------------------------------------------------------------+ - [multilabel_classify.py:837:log_training_args] 2025-06-03 20:22:22,368 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:96:log_section] 2025-06-03 20:22:22,368 - INFO - + โœจ MODEL TRAINING + - [multilabel_classify.py:97:log_section] 2025-06-03 20:22:22,368 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:100:log_section] 2025-06-03 20:22:22,368 - INFO - ๐Ÿ‹๏ธ Starting model training process... - [multilabel_classify.py:3340:main] 2025-06-03 20:22:22,411 - INFO - We are registering the tokenizer deb101/mistral-7b-instruct-v0.3-mimic4-adapt in Custom Trainer - [multilabel_classify.py:1941:__init__] 2025-06-03 20:22:22,667 - INFO - ๐Ÿš€ Starting Training... - [multilabel_classify.py:1595:on_train_begin] 2025-06-03 20:22:43,213 - INFO - ๐Ÿš‚ Training Metrics (Step 10) ๐Ÿš‚ +---------------+---------+ | Metric | Value | +===============+=========+ | loss | 0.624 | +---------------+---------+ | grad_norm | 6.71916 | +---------------+---------+ | learning_rate | 2e-06 | +---------------+---------+ | epoch | 0.56338 | +---------------+---------+ - [multilabel_classify.py:1789:on_log] 2025-06-03 20:22:57,293 - INFO - Removing 'token_type_ids' from eval_dataset as they are not needed. - [multilabel_classify.py:1953:evaluate] 2025-06-03 20:23:32,597 - INFO - ๐Ÿ” Evaluation Metrics ๐Ÿ” +-------------------------------+----------+ | Metric | Value | +===============================+==========+ | eval_f1_micro | 0 | +-------------------------------+----------+ | eval_f1_macro | 0 | +-------------------------------+----------+ | eval_precision_at_5 | 0.005882 | +-------------------------------+----------+ | eval_recall_at_5 | 0.001306 | +-------------------------------+----------+ | eval_precision_at_8 | 0.011029 | +-------------------------------+----------+ | eval_recall_at_8 | 0.003704 | +-------------------------------+----------+ | eval_precision_at_15 | 0.014706 | +-------------------------------+----------+ | eval_recall_at_15 | 0.011093 | +-------------------------------+----------+ | eval_rare_f1_micro | 0 | +-------------------------------+----------+ | eval_rare_f1_macro | 0 | +-------------------------------+----------+ | eval_rare_precision | 0 | +-------------------------------+----------+ | eval_rare_recall | 0 | +-------------------------------+----------+ | eval_rare_precision_at_5 | 0.010294 | +-------------------------------+----------+ | eval_rare_recall_at_5 | 0.002439 | +-------------------------------+----------+ | eval_rare_precision_at_8 | 0.012868 | +-------------------------------+----------+ | eval_rare_recall_at_8 | 0.00592 | +-------------------------------+----------+ | eval_rare_precision_at_15 | 0.014216 | +-------------------------------+----------+ | eval_rare_recall_at_15 | 0.011398 | +-------------------------------+----------+ | eval_not_rare_f1_micro | 0.595588 | +-------------------------------+----------+ | eval_not_rare_f1_macro | 0.373272 | +-------------------------------+----------+ | eval_not_rare_precision | 0.595588 | +-------------------------------+----------+ | eval_not_rare_recall | 0.595588 | +-------------------------------+----------+ | eval_not_rare_precision_at_5 | 0.080882 | +-------------------------------+----------+ | eval_not_rare_recall_at_5 | 0.404412 | +-------------------------------+----------+ | eval_not_rare_precision_at_8 | 0.050551 | +-------------------------------+----------+ | eval_not_rare_recall_at_8 | 0.404412 | +-------------------------------+----------+ | eval_not_rare_precision_at_15 | 0.026961 | +-------------------------------+----------+ | eval_not_rare_recall_at_15 | 0.404412 | +-------------------------------+----------+ | eval_loss | 0.221656 | +-------------------------------+----------+ - [multilabel_classify.py:1808:on_evaluate] 2025-06-03 20:23:36,490 - INFO - ๐Ÿ’พ Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-18 - [multilabel_classify.py:2046:_save] 2025-06-03 20:23:36,493 - INFO - โš™๏ธ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-18 - [multilabel_classify.py:2051:_save] 2025-06-03 20:23:36,494 - INFO - ๐Ÿ“‹ Saved files in ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-18: +---------+--------------------+------------+ | Index | Saved File | Size | +=========+====================+============+ | 1 | training_args.bin | 0.01 MB | +---------+--------------------+------------+ | 2 | optimizer.pt | 642.30 MB | +---------+--------------------+------------+ | 3 | model.safetensors | 4267.74 MB | +---------+--------------------+------------+ | 4 | scaler.pt | 0.00 MB | +---------+--------------------+------------+ | 5 | config.json | 0.04 MB | +---------+--------------------+------------+ | 6 | scheduler.pt | 0.00 MB | +---------+--------------------+------------+ | 7 | trainer_state.json | 0.00 MB | +---------+--------------------+------------+ | 8 | rng_state.pth | 0.01 MB | +---------+--------------------+------------+ - [multilabel_classify.py:2068:_save] 2025-06-03 20:23:42,667 - INFO - ๐Ÿš‚ Training Metrics (Step 20) ๐Ÿš‚ +---------------+---------+ | Metric | Value | +===============+=========+ | loss | 0.3386 | +---------------+---------+ | grad_norm | 2.49967 | +---------------+---------+ | learning_rate | 4e-06 | +---------------+---------+ | epoch | 1.11268 | +---------------+---------+ - [multilabel_classify.py:1789:on_log] 2025-06-03 20:24:00,820 - INFO - ๐Ÿš‚ Training Metrics (Step 30) ๐Ÿš‚ +---------------+----------+ | Metric | Value | +===============+==========+ | loss | 0.1249 | +---------------+----------+ | grad_norm | 0.258035 | +---------------+----------+ | learning_rate | 6e-06 | +---------------+----------+ | epoch | 1.67606 | +---------------+----------+ - [multilabel_classify.py:1789:on_log] 2025-06-03 20:24:11,212 - INFO - Removing 'token_type_ids' from eval_dataset as they are not needed. - [multilabel_classify.py:1953:evaluate] 2025-06-03 20:24:46,604 - INFO - ๐Ÿ” Evaluation Metrics ๐Ÿ” +-------------------------------+----------+ | Metric | Value | +===============================+==========+ | eval_f1_micro | 0 | +-------------------------------+----------+ | eval_f1_macro | 0 | +-------------------------------+----------+ | eval_precision_at_5 | 0.008824 | +-------------------------------+----------+ | eval_recall_at_5 | 0.001777 | +-------------------------------+----------+ | eval_precision_at_8 | 0.011029 | +-------------------------------+----------+ | eval_recall_at_8 | 0.003763 | +-------------------------------+----------+ | eval_precision_at_15 | 0.015196 | +-------------------------------+----------+ | eval_recall_at_15 | 0.010958 | +-------------------------------+----------+ | eval_rare_f1_micro | 0 | +-------------------------------+----------+ | eval_rare_f1_macro | 0 | +-------------------------------+----------+ | eval_rare_precision | 0 | +-------------------------------+----------+ | eval_rare_recall | 0 | +-------------------------------+----------+ | eval_rare_precision_at_5 | 0.008824 | +-------------------------------+----------+ | eval_rare_recall_at_5 | 0.001785 | +-------------------------------+----------+ | eval_rare_precision_at_8 | 0.011029 | +-------------------------------+----------+ | eval_rare_recall_at_8 | 0.004688 | +-------------------------------+----------+ | eval_rare_precision_at_15 | 0.014216 | +-------------------------------+----------+ | eval_rare_recall_at_15 | 0.012365 | +-------------------------------+----------+ | eval_not_rare_f1_micro | 0.595588 | +-------------------------------+----------+ | eval_not_rare_f1_macro | 0.373272 | +-------------------------------+----------+ | eval_not_rare_precision | 0.595588 | +-------------------------------+----------+ | eval_not_rare_recall | 0.595588 | +-------------------------------+----------+ | eval_not_rare_precision_at_5 | 0.080882 | +-------------------------------+----------+ | eval_not_rare_recall_at_5 | 0.404412 | +-------------------------------+----------+ | eval_not_rare_precision_at_8 | 0.050551 | +-------------------------------+----------+ | eval_not_rare_recall_at_8 | 0.404412 | +-------------------------------+----------+ | eval_not_rare_precision_at_15 | 0.026961 | +-------------------------------+----------+ | eval_not_rare_recall_at_15 | 0.404412 | +-------------------------------+----------+ | eval_loss | 0.121922 | +-------------------------------+----------+ - [multilabel_classify.py:1808:on_evaluate] 2025-06-03 20:24:50,330 - INFO - ๐Ÿ’พ Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-36 - [multilabel_classify.py:2046:_save] 2025-06-03 20:24:50,333 - INFO - โš™๏ธ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-36 - [multilabel_classify.py:2051:_save] 2025-06-03 20:24:50,334 - INFO - ๐Ÿ“‹ Saved files in ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-36: +---------+--------------------+------------+ | Index | Saved File | Size | +=========+====================+============+ | 1 | training_args.bin | 0.01 MB | +---------+--------------------+------------+ | 2 | optimizer.pt | 642.30 MB | +---------+--------------------+------------+ | 3 | model.safetensors | 4267.74 MB | +---------+--------------------+------------+ | 4 | scaler.pt | 0.00 MB | +---------+--------------------+------------+ | 5 | config.json | 0.04 MB | +---------+--------------------+------------+ | 6 | scheduler.pt | 0.00 MB | +---------+--------------------+------------+ | 7 | trainer_state.json | 0.00 MB | +---------+--------------------+------------+ | 8 | rng_state.pth | 0.01 MB | +---------+--------------------+------------+ - [multilabel_classify.py:2068:_save] 2025-06-03 20:25:00,280 - INFO - ๐Ÿš‚ Training Metrics (Step 40) ๐Ÿš‚ +---------------+----------+ | Metric | Value | +===============+==========+ | loss | 0.1098 | +---------------+----------+ | grad_norm | 0.470693 | +---------------+----------+ | learning_rate | 8e-06 | +---------------+----------+ | epoch | 2.22535 | +---------------+----------+ - [multilabel_classify.py:1789:on_log] 2025-06-03 20:25:18,391 - INFO - ๐Ÿš‚ Training Metrics (Step 50) ๐Ÿš‚ +---------------+----------+ | Metric | Value | +===============+==========+ | loss | 0.1086 | +---------------+----------+ | grad_norm | 0.067761 | +---------------+----------+ | learning_rate | 1e-05 | +---------------+----------+ | epoch | 2.78873 | +---------------+----------+ - [multilabel_classify.py:1789:on_log] 2025-06-03 20:25:25,137 - INFO - Removing 'token_type_ids' from eval_dataset as they are not needed. - [multilabel_classify.py:1953:evaluate] 2025-06-03 20:26:00,429 - INFO - ๐Ÿ” Evaluation Metrics ๐Ÿ” +-------------------------------+----------+ | Metric | Value | +===============================+==========+ | eval_f1_micro | 0 | +-------------------------------+----------+ | eval_f1_macro | 0 | +-------------------------------+----------+ | eval_precision_at_5 | 0.047059 | +-------------------------------+----------+ | eval_recall_at_5 | 0.012123 | +-------------------------------+----------+ | eval_precision_at_8 | 0.050551 | +-------------------------------+----------+ | eval_recall_at_8 | 0.02375 | +-------------------------------+----------+ | eval_precision_at_15 | 0.045588 | +-------------------------------+----------+ | eval_recall_at_15 | 0.045718 | +-------------------------------+----------+ | eval_rare_f1_micro | 0 | +-------------------------------+----------+ | eval_rare_f1_macro | 0 | +-------------------------------+----------+ | eval_rare_precision | 0 | +-------------------------------+----------+ | eval_rare_recall | 0 | +-------------------------------+----------+ | eval_rare_precision_at_5 | 0.036765 | +-------------------------------+----------+ | eval_rare_recall_at_5 | 0.010554 | +-------------------------------+----------+ | eval_rare_precision_at_8 | 0.04136 | +-------------------------------+----------+ | eval_rare_recall_at_8 | 0.019845 | +-------------------------------+----------+ | eval_rare_precision_at_15 | 0.036765 | +-------------------------------+----------+ | eval_rare_recall_at_15 | 0.037016 | +-------------------------------+----------+ | eval_not_rare_f1_micro | 0.595588 | +-------------------------------+----------+ | eval_not_rare_f1_macro | 0.373272 | +-------------------------------+----------+ | eval_not_rare_precision | 0.595588 | +-------------------------------+----------+ | eval_not_rare_recall | 0.595588 | +-------------------------------+----------+ | eval_not_rare_precision_at_5 | 0.080882 | +-------------------------------+----------+ | eval_not_rare_recall_at_5 | 0.404412 | +-------------------------------+----------+ | eval_not_rare_precision_at_8 | 0.050551 | +-------------------------------+----------+ | eval_not_rare_recall_at_8 | 0.404412 | +-------------------------------+----------+ | eval_not_rare_precision_at_15 | 0.026961 | +-------------------------------+----------+ | eval_not_rare_recall_at_15 | 0.404412 | +-------------------------------+----------+ | eval_loss | 0.107352 | +-------------------------------+----------+ - [multilabel_classify.py:1808:on_evaluate] 2025-06-03 20:26:01,710 - INFO - ๐Ÿ’พ Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-54 - [multilabel_classify.py:2046:_save] 2025-06-03 20:26:01,713 - INFO - โš™๏ธ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-54 - [multilabel_classify.py:2051:_save] 2025-06-03 20:26:01,714 - INFO - ๐Ÿ“‹ Saved files in ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-54: +---------+-------------------+------------+ | Index | Saved File | Size | +=========+===================+============+ | 1 | training_args.bin | 0.01 MB | +---------+-------------------+------------+ | 2 | model.safetensors | 4267.74 MB | +---------+-------------------+------------+ | 3 | config.json | 0.04 MB | +---------+-------------------+------------+ - [multilabel_classify.py:2068:_save] 2025-06-03 20:26:15,035 - INFO - ๐Ÿš‚ Training Metrics (Step 60) ๐Ÿš‚ +---------------+----------+ | Metric | Value | +===============+==========+ | loss | 0.1033 | +---------------+----------+ | grad_norm | 0.206313 | +---------------+----------+ | learning_rate | 1.2e-05 | +---------------+----------+ | epoch | 3.33803 | +---------------+----------+ - [multilabel_classify.py:1789:on_log] 2025-06-03 20:26:30,807 - INFO - ๐Ÿ’พ Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-68 - [multilabel_classify.py:2046:_save] 2025-06-03 20:26:30,810 - INFO - โš™๏ธ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-68 - [multilabel_classify.py:2051:_save] 2025-06-03 20:26:30,811 - INFO - ๐Ÿ“‹ Saved files in ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-68: +---------+-------------------+------------+ | Index | Saved File | Size | +=========+===================+============+ | 1 | training_args.bin | 0.01 MB | +---------+-------------------+------------+ | 2 | model.safetensors | 4267.74 MB | +---------+-------------------+------------+ | 3 | config.json | 0.04 MB | +---------+-------------------+------------+ - [multilabel_classify.py:2068:_save] 2025-06-03 20:26:31,136 - INFO - Removing 'token_type_ids' from eval_dataset as they are not needed. - [multilabel_classify.py:1953:evaluate] 2025-06-03 20:27:06,536 - INFO - ๐Ÿ” Evaluation Metrics ๐Ÿ” +-------------------------------+----------+ | Metric | Value | +===============================+==========+ | eval_f1_micro | 0 | +-------------------------------+----------+ | eval_f1_macro | 0 | +-------------------------------+----------+ | eval_precision_at_5 | 0.167647 | +-------------------------------+----------+ | eval_recall_at_5 | 0.059694 | +-------------------------------+----------+ | eval_precision_at_8 | 0.137868 | +-------------------------------+----------+ | eval_recall_at_8 | 0.077015 | +-------------------------------+----------+ | eval_precision_at_15 | 0.092647 | +-------------------------------+----------+ | eval_recall_at_15 | 0.099223 | +-------------------------------+----------+ | eval_rare_f1_micro | 0 | +-------------------------------+----------+ | eval_rare_f1_macro | 0 | +-------------------------------+----------+ | eval_rare_precision | 0 | +-------------------------------+----------+ | eval_rare_recall | 0 | +-------------------------------+----------+ | eval_rare_precision_at_5 | 0.127941 | +-------------------------------+----------+ | eval_rare_recall_at_5 | 0.046166 | +-------------------------------+----------+ | eval_rare_precision_at_8 | 0.09375 | +-------------------------------+----------+ | eval_rare_recall_at_8 | 0.052229 | +-------------------------------+----------+ | eval_rare_precision_at_15 | 0.064216 | +-------------------------------+----------+ | eval_rare_recall_at_15 | 0.069951 | +-------------------------------+----------+ | eval_not_rare_f1_micro | 0.595588 | +-------------------------------+----------+ | eval_not_rare_f1_macro | 0.373272 | +-------------------------------+----------+ | eval_not_rare_precision | 0.595588 | +-------------------------------+----------+ | eval_not_rare_recall | 0.595588 | +-------------------------------+----------+ | eval_not_rare_precision_at_5 | 0.080882 | +-------------------------------+----------+ | eval_not_rare_recall_at_5 | 0.404412 | +-------------------------------+----------+ | eval_not_rare_precision_at_8 | 0.050551 | +-------------------------------+----------+ | eval_not_rare_recall_at_8 | 0.404412 | +-------------------------------+----------+ | eval_not_rare_precision_at_15 | 0.026961 | +-------------------------------+----------+ | eval_not_rare_recall_at_15 | 0.404412 | +-------------------------------+----------+ | eval_loss | 0.10503 | +-------------------------------+----------+ - [multilabel_classify.py:1808:on_evaluate] 2025-06-03 20:27:10,158 - INFO - ๐Ÿ’พ Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-68 - [multilabel_classify.py:2046:_save] 2025-06-03 20:27:10,161 - INFO - โš™๏ธ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-68 - [multilabel_classify.py:2051:_save] 2025-06-03 20:27:10,163 - INFO - ๐Ÿ“‹ Saved files in ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-68: +---------+--------------------+------------+ | Index | Saved File | Size | +=========+====================+============+ | 1 | training_args.bin | 0.01 MB | +---------+--------------------+------------+ | 2 | optimizer.pt | 642.30 MB | +---------+--------------------+------------+ | 3 | model.safetensors | 4267.74 MB | +---------+--------------------+------------+ | 4 | scaler.pt | 0.00 MB | +---------+--------------------+------------+ | 5 | config.json | 0.04 MB | +---------+--------------------+------------+ | 6 | scheduler.pt | 0.00 MB | +---------+--------------------+------------+ | 7 | trainer_state.json | 0.01 MB | +---------+--------------------+------------+ | 8 | rng_state.pth | 0.01 MB | +---------+--------------------+------------+ - [multilabel_classify.py:2068:_save] 2025-06-03 20:27:10,994 - INFO - ๐Ÿ“‚ Loading best model from ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-68 - [multilabel_classify.py:2120:_load_best_model] 2025-06-03 20:27:10,994 - INFO - ๐Ÿ–ฅ๏ธ Model is on device: cuda:0 - [multilabel_classify.py:2130:_load_best_model] 2025-06-03 20:27:11,049 - INFO - ๐Ÿ”‘ Key order comparison: +---------+--------------------------------------------+--------------------------------------------------------------------------------------+ | Index | Saved state_dict Keys | Model state_dict Keys | +=========+============================================+======================================================================================+ | 1 | attention.in_proj_bias | boost_mul | +---------+--------------------------------------------+--------------------------------------------------------------------------------------+ | 2 | attention.in_proj_weight | boost_add | +---------+--------------------------------------------+--------------------------------------------------------------------------------------+ | 3 | attention.out_proj.bias | base_model.base_model.model.model.embed_tokens.weight | +---------+--------------------------------------------+--------------------------------------------------------------------------------------+ | 4 | attention.out_proj.weight | base_model.base_model.model.model.layers.0.self_attn.q_proj.base_layer.weight | +---------+--------------------------------------------+--------------------------------------------------------------------------------------+ | 5 | base_model.base_model.model.lm_head.weight | base_model.base_model.model.model.layers.0.self_attn.q_proj.base_layer.weight.absmax | +---------+--------------------------------------------+--------------------------------------------------------------------------------------+ - [multilabel_classify.py:2154:_load_best_model] 2025-06-03 20:27:12,079 - INFO - โœ… Loaded best model weights from ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/checkpoint-68/model.safetensors - [multilabel_classify.py:2171:_load_best_model] 2025-06-03 20:27:12,105 - INFO - โœ”๏ธ Weight for boost_mul matches between saved and loaded state_dict - [multilabel_classify.py:2183:_load_best_model] 2025-06-03 20:27:12,129 - INFO - โœ”๏ธ Weight for boost_add matches between saved and loaded state_dict - [multilabel_classify.py:2183:_load_best_model] 2025-06-03 20:27:12,152 - INFO - ๐Ÿš‚ Training Metrics (Step 68) ๐Ÿš‚ +--------------------------+---------+ | Metric | Value | +==========================+=========+ | train_runtime | 289.486 | +--------------------------+---------+ | train_samples_per_second | 7.835 | +--------------------------+---------+ | train_steps_per_second | 0.235 | +--------------------------+---------+ | total_flos | 0 | +--------------------------+---------+ | train_loss | 0.21949 | +--------------------------+---------+ | epoch | 3.78873 | +--------------------------+---------+ - [multilabel_classify.py:1789:on_log] 2025-06-03 20:27:12,152 - INFO - โœจ Training Completed! โœจ - [multilabel_classify.py:1662:on_train_end] 2025-06-03 20:27:12,224 - INFO - ๐Ÿ“Š Training loss plot saved as '../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/train_loss_plot.png' - [multilabel_classify.py:1858:on_train_end] 2025-06-03 20:27:12,286 - INFO - ๐Ÿ“Š Evaluation loss plot saved as '../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/eval_loss_plot.png' - [multilabel_classify.py:1872:on_train_end] 2025-06-03 20:27:12,348 - INFO - ๐Ÿ“Š Evaluation metric plot saved as '../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b/eval_precision_at_15_plot.png' - [multilabel_classify.py:1893:on_train_end] 2025-06-03 20:27:12,348 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:96:log_section] 2025-06-03 20:27:12,348 - INFO - + โœจ MODEL SAVING + - [multilabel_classify.py:97:log_section] 2025-06-03 20:27:12,348 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [multilabel_classify.py:100:log_section] 2025-06-03 20:27:12,349 - INFO - ๐Ÿ’พ Saving trained model and pushing to Hugging Face Hub... - [multilabel_classify.py:3354:main] 2025-06-03 20:27:12,349 - INFO - ๐Ÿ“ Creating/using output directory: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b - [multilabel_classify.py:2646:save_and_push] 2025-06-03 20:27:13,614 - INFO - ๐Ÿ’พ Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b - [multilabel_classify.py:2046:_save] 2025-06-03 20:27:13,617 - INFO - โš™๏ธ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b - [multilabel_classify.py:2051:_save] 2025-06-03 20:27:13,618 - INFO - ๐Ÿ“‹ Saved files in ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b: +---------+-------------------------------+------------+ | Index | Saved File | Size | +=========+===============================+============+ | 1 | eval_loss_plot.png | 0.03 MB | +---------+-------------------------------+------------+ | 2 | training_args.bin | 0.01 MB | +---------+-------------------------------+------------+ | 3 | model.safetensors | 4267.74 MB | +---------+-------------------------------+------------+ | 4 | config.json | 0.04 MB | +---------+-------------------------------+------------+ | 5 | train_loss_plot.png | 0.02 MB | +---------+-------------------------------+------------+ | 6 | eval_precision_at_15_plot.png | 0.03 MB | +---------+-------------------------------+------------+ - [multilabel_classify.py:2068:_save] 2025-06-03 20:27:17,349 - INFO - ๐Ÿ’พ Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b - [multilabel_classify.py:2046:_save] 2025-06-03 20:27:17,352 - INFO - โš™๏ธ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b - [multilabel_classify.py:2051:_save] 2025-06-03 20:27:17,354 - INFO - ๐Ÿ“‹ Saved files in ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b: +---------+-------------------------------+------------+ | Index | Saved File | Size | +=========+===============================+============+ | 1 | eval_loss_plot.png | 0.03 MB | +---------+-------------------------------+------------+ | 2 | training_args.bin | 0.01 MB | +---------+-------------------------------+------------+ | 3 | model.safetensors | 4267.74 MB | +---------+-------------------------------+------------+ | 4 | config.json | 0.04 MB | +---------+-------------------------------+------------+ | 5 | train_loss_plot.png | 0.02 MB | +---------+-------------------------------+------------+ | 6 | eval_precision_at_15_plot.png | 0.03 MB | +---------+-------------------------------+------------+ - [multilabel_classify.py:2068:_save] 2025-06-03 20:28:44,815 - INFO - ๐Ÿ’พ Model saved to: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b - [multilabel_classify.py:2650:save_and_push] 2025-06-03 20:28:44,845 - INFO - ๐Ÿ–Œ๏ธ Tokenizer saved to: ../tmp/MIMIC4_DEMO/mimic4_classify_mistral7b - [multilabel_classify.py:2654:save_and_push]