mistral-7b-instruct-v0.3-mimic4-adapt-l2r / training_l2r_log_2025-06-09_20-58-40.log
deb101's picture
Trained L2R token ranking model on MIMIC-IV
161f7f2 verified
2025-06-09 20:58:40,897 - INFO - 📝 Logging initialized. Log file created at: ../tmp/logs/training_l2r_log_2025-06-09_20-58-40.log - [learning2rank.py:288:setup_logging]
2025-06-09 20:58:40,897 - INFO - ================================================================================ - [learning2rank.py:110:log_section]
2025-06-09 20:58:40,897 - INFO - = 📌 INITIALIZING TRAINING ENVIRONMENT = - [learning2rank.py:111:log_section]
2025-06-09 20:58:40,897 - INFO - ================================================================================ - [learning2rank.py:114:log_section]
2025-06-09 20:58:40,897 - INFO - 🚀 Setting up data paths and environment variables... - [learning2rank.py:3916:main]
2025-06-09 20:58:40,898 - INFO - 🛠️ Command-line Arguments: - [learning2rank.py:304:print_args]
2025-06-09 20:58:40,898 - INFO -
🔹 output_dir: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b
🔹 source_url: XURLs.MIMIC4_DEMO
🔹 data: mimic4_icd10_full
🔹 data_l2r_fname_prefix: mimic4_icd10
🔹 logfile: training_l2r_log
🔹 base_dir: ../tmp/MIMIC4_DEMO
🔹 l2r_boot_dir: ../tmp/MIMIC4_DEMO/mimic4_l2rboot_mistral7b
🔹 hub_model_id: deb101/mistral-7b-instruct-v0.3-mimic4-adapt
🔹 model_name: mistralai/Mistral-7B-Instruct-v0.3
🔹 max_length: 512
🔹 do_fresh_training: True
🔹 load_from_checkpoint: False
🔹 task: l2r
🔹 num_train_epochs: 4
🔹 metric_for_best_model: ndcg@25
🔹 learning_rate: 0.0001
🔹 warmup_steps: 0
🔹 generate_report: False
🔹 logfile_path: ../tmp/logs/training_l2r_log_2025-06-09_20-58-40.log
🔹 source: /home/ubuntu/.xcube/data/mimic4_demo - [learning2rank.py:305:print_args]
2025-06-09 20:58:40,898 - INFO - ➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖➖ - [learning2rank.py:306:print_args]
2025-06-09 20:58:40,898 - INFO - 📁 Using L2R boot directory: ../tmp/MIMIC4_DEMO/mimic4_l2rboot_mistral7b - [learning2rank.py:3930:main]
2025-06-09 20:58:40,898 - INFO - 📂 Using output directory: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b - [learning2rank.py:3932:main]
2025-06-09 20:58:40,898 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:110:log_section]
2025-06-09 20:58:40,898 - INFO - + LOADING DATASETS + - [learning2rank.py:111:log_section]
2025-06-09 20:58:40,898 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:114:log_section]
2025-06-09 20:58:40,898 - INFO - 📊 Loading main dataset and L2R dataset... - [learning2rank.py:3935:main]
2025-06-09 20:58:40,898 - INFO - 📂 Loading main data from: /home/ubuntu/.xcube/data/mimic4_demo/mimic4_icd10_full.csv - [learning2rank.py:334:get_data]
2025-06-09 20:58:49,152 - INFO - Successfully loaded main data: 122279 rows - [learning2rank.py:347:get_data]
2025-06-09 20:58:49,152 - INFO - 📂 Loading L2R data from: ../tmp/MIMIC4_DEMO/mimic4_l2rboot_mistral7b/mimic4_icd10_tok_rank_per_lbl.ft - [learning2rank.py:353:get_data]
2025-06-09 20:58:49,152 - INFO - 📂 Loading L2R tokens from: ../tmp/MIMIC4_DEMO/mimic4_l2rboot_mistral7b/mimic4_icd10_tok.ft - [learning2rank.py:356:get_data]
2025-06-09 20:58:49,152 - INFO - 📂 Loading L2R labels from: ../tmp/MIMIC4_DEMO/mimic4_l2rboot_mistral7b/mimic4_icd10_lbl.ft - [learning2rank.py:359:get_data]
2025-06-09 20:58:50,870 - INFO - Successfully loaded L2R data: 260243456 rows - [learning2rank.py:366:get_data]
2025-06-09 20:58:50,870 - INFO - Successfully loaded L2R tokens: 32768 tokens - [learning2rank.py:367:get_data]
2025-06-09 20:58:50,870 - INFO - Successfully loaded L2R labels: 7942 rows - [learning2rank.py:368:get_data]
2025-06-09 20:58:50,870 - INFO - 🔄 Total data loaded: 122279 main rows, 260243456 L2R rows - [learning2rank.py:375:get_data]
2025-06-09 20:58:50,870 - INFO - Successfully loaded both datasets: - [learning2rank.py:399:load_datasets]
2025-06-09 20:58:50,870 - INFO - - 📄 Main dataset: 122279 records - [learning2rank.py:400:load_datasets]
2025-06-09 20:58:50,870 - INFO - - 📄 L2R dataset: 260243456 records - [learning2rank.py:401:load_datasets]
2025-06-09 20:58:50,870 - INFO - - 🔤 Tokens: 32768 items - [learning2rank.py:402:load_datasets]
2025-06-09 20:58:50,871 - INFO - - 🏷️ Labels: 7942 items - [learning2rank.py:403:load_datasets]
2025-06-09 20:58:50,878 - INFO - Data loading completed successfully - [learning2rank.py:411:load_datasets]
2025-06-09 20:58:52,154 - INFO - Starting quantization of ranks for DataFrame with 260243456 rows. Containing 32768 unique tokens & 7942 unique labels - [learning2rank.py:530:quantize_l2r_data]
2025-06-09 20:58:52,155 - INFO - 🔄 Quantizing those 32768 unique token ranks into 101 quantization levels for each label - [learning2rank.py:555:quantize_l2r_data]
2025-06-09 20:59:38,615 - INFO - Completed quantization: Produced tensor of shape torch.Size([7942, 32768, 4]) with 101 quantization levels per label - [learning2rank.py:609:quantize_l2r_data]
2025-06-09 20:59:38,642 - WARNING - Label 6031: Only 1 tokens with top relevance score (need 50) - [learning2rank.py:725:test_scored_tokens]
2025-06-09 20:59:38,648 - WARNING - Label 3729: Only 1 tokens with top relevance score (need 50) - [learning2rank.py:725:test_scored_tokens]
2025-06-09 20:59:38,651 - WARNING - Label 4801: Only 1 tokens with top relevance score (need 50) - [learning2rank.py:725:test_scored_tokens]
2025-06-09 20:59:38,655 - WARNING - Label 3021: Only 1 tokens with top relevance score (need 50) - [learning2rank.py:725:test_scored_tokens]
2025-06-09 20:59:38,659 - WARNING - Label 6454: Only 1 tokens with top relevance score (need 50) - [learning2rank.py:725:test_scored_tokens]
2025-06-09 20:59:38,687 - INFO - ******************************************************************************** - [learning2rank.py:110:log_section]
2025-06-09 20:59:38,687 - INFO - * 🌟 STARTING LEARNING TO RANK MODEL TRAINING * - [learning2rank.py:111:log_section]
2025-06-09 20:59:38,687 - INFO - ******************************************************************************** - [learning2rank.py:114:log_section]
2025-06-09 20:59:38,687 - INFO - 🔐 Loaded authentication token from environment - [learning2rank.py:3964:main]
2025-06-09 20:59:38,687 - INFO - 🏷️ Hub Model ID for this Learning to Rank task: deb101/mistral-7b-instruct-v0.3-mimic4-adapt-l2r - [learning2rank.py:3968:main]
2025-06-09 20:59:38,687 - INFO - -------------------------------------------------------------------------------- - [learning2rank.py:110:log_section]
2025-06-09 20:59:38,687 - INFO - - 📋 MODEL EXISTENCE CHECK - - [learning2rank.py:111:log_section]
2025-06-09 20:59:38,687 - INFO - -------------------------------------------------------------------------------- - [learning2rank.py:114:log_section]
2025-06-09 20:59:38,687 - INFO - 🔍 Checking model existence locally and on Hugging Face Hub... - [learning2rank.py:431:check_model_existence]
2025-06-09 20:59:38,687 - INFO - Model not found locally at: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b - [learning2rank.py:438:check_model_existence]
2025-06-09 20:59:38,739 - INFO - Model exists on Hugging Face Hub with ID: deb101/mistral-7b-instruct-v0.3-mimic4-adapt-l2r - [learning2rank.py:450:check_model_existence]
2025-06-09 20:59:38,739 - INFO - 📁 Model exists either locally or on Hub - [learning2rank.py:476:check_model_existence]
2025-06-09 20:59:38,739 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:110:log_section]
2025-06-09 20:59:38,739 - INFO - + STARTING FRESH TRAINING + - [learning2rank.py:111:log_section]
2025-06-09 20:59:38,739 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:114:log_section]
2025-06-09 20:59:38,739 - INFO - 🔄 Starting fresh training (either forced or model not found)... - [learning2rank.py:3981:main]
2025-06-09 20:59:38,805 - 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]
2025-06-09 20:59:38,805 - INFO - 🔑 Successfully authenticated with Hugging Face Hub - [learning2rank.py:312:authenticate_hf]
2025-06-09 20:59:38,805 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:110:log_section]
2025-06-09 20:59:38,805 - INFO - + LOADING BASE MODEL + - [learning2rank.py:111:log_section]
2025-06-09 20:59:38,805 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:114:log_section]
2025-06-09 20:59:38,805 - INFO - 📥 Loading pretrained model and tokenizer... - [learning2rank.py:4023:main]
2025-06-09 20:59:38,805 - INFO - 🚀 Starting model and tokenizer loading process... - [learning2rank.py:940:load_base_model_and_tokenizer]
2025-06-09 20:59:38,806 - INFO - 📊 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"
}
- [learning2rank.py:949:load_base_model_and_tokenizer]
2025-06-09 20:59:38,806 - INFO - 🔤 Loading tokenizer for model: mistralai/Mistral-7B-Instruct-v0.3... - [learning2rank.py:953:load_base_model_and_tokenizer]
2025-06-09 20:59:39,085 - INFO - 📝 Setting pad token to eos token... - [learning2rank.py:957:load_base_model_and_tokenizer]
2025-06-09 20:59:39,085 - INFO - 🧠 Loading base model with quantization... - [learning2rank.py:965:load_base_model_and_tokenizer]
2025-06-09 20:59:39,590 - 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]
2025-06-09 20:59:44,833 - INFO - 🔧 Setting up default LoRA configuration... - [learning2rank.py:988:load_base_model_and_tokenizer]
2025-06-09 20:59:44,833 - INFO - 🔍 LoRA config: r=16, alpha=32, targets={'q_proj', 'v_proj', 'k_proj', 'o_proj'}, dropout=0.05 - [learning2rank.py:1011:load_base_model_and_tokenizer]
2025-06-09 20:59:44,833 - INFO - 🧩 Applying LoRA adapters to model... - [learning2rank.py:1018:load_base_model_and_tokenizer]
2025-06-09 20:59:45,006 - INFO - 📊 trainable params: 13,631,488 || all params: 7,261,655,040 || trainable%: 0.1877 - [learning2rank.py:136:log_print_output]
2025-06-09 20:59:45,006 - INFO - Model and tokenizer successfully loaded! - [learning2rank.py:1025:load_base_model_and_tokenizer]
2025-06-09 20:59:45,006 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:110:log_section]
2025-06-09 20:59:45,006 - INFO - + DATA PREPROCESSING + - [learning2rank.py:111:log_section]
2025-06-09 20:59:45,006 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:114:log_section]
2025-06-09 20:59:45,006 - INFO - 🔄 Loading and preprocessing training data... - [learning2rank.py:4031:main]
2025-06-09 20:59:46,078 - INFO - 🔍 Verifying uniqueness of token IDs per label in scored_tokens... - [learning2rank.py:1303:preprocess_data]
2025-06-09 20:59:46,920 - INFO - All labels have unique token IDs in scored_tokens. - [learning2rank.py:1315:preprocess_data]
2025-06-09 20:59:46,920 - INFO - 🚀 Building a 2D lookup table for efficient token-to-relevance mapping across all labels! 🚀 - [learning2rank.py:1318:preprocess_data]
2025-06-09 20:59:46,920 - INFO - 🔢 Total labels = 7942 - [learning2rank.py:1321:preprocess_data]
2025-06-09 20:59:46,920 - INFO - 🔍 Precomputing token indices and corresponding relevance_values for each label... - [learning2rank.py:1322:preprocess_data]
2025-06-09 20:59:47,148 - INFO - 📊 Lookup table dimensions: 32768 vocabulary size × 7942 labels - [learning2rank.py:1329:preprocess_data]
2025-06-09 20:59:47,148 - INFO - This approach eliminates token comparison broadcasting and provides O(1) lookup time for relevance scores! - [learning2rank.py:1332:preprocess_data]
2025-06-09 20:59:47,148 - INFO - 🧮 Processing relevance calculations vectorized for maximum speed 🔥 - [learning2rank.py:1335:preprocess_data]
2025-06-09 20:59:47,315 - INFO - 🔍 Verifying token mappings with 10 samples... - [learning2rank.py:1368:verify_token_mappings]
2025-06-09 20:59:47,482 - INFO - Token mappings verification completed successfully! 🎉 - [learning2rank.py:1459:verify_token_mappings]
2025-06-09 20:59:47,487 - INFO - 🔄 Processing dataset with map... - [learning2rank.py:1523:preprocess_data]
2025-06-09 20:59:47,804 - INFO - Dataset built in 0h 0m 0.32s - [learning2rank.py:1546:preprocess_data]
2025-06-09 20:59:47,816 - INFO - The size of Training set: 173 🏋️ (Training Data) - [learning2rank.py:1577:preprocess_data]
2025-06-09 20:59:47,816 - INFO - The size of Evaluation set: 35 🔍 (Test Data) - [learning2rank.py:1578:preprocess_data]
2025-06-09 20:59:47,816 - INFO - 🚀 Created HuggingFace Dataset with 208 samples, 7942 labels - [learning2rank.py:1586:preprocess_data]
2025-06-09 20:59:47,817 - INFO - 🏷️ Number of unique ICD-10 codes: 7942 - [learning2rank.py:4044:main]
2025-06-09 20:59:47,817 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:110:log_section]
2025-06-09 20:59:47,817 - INFO - + MODEL INITIALIZATION + - [learning2rank.py:111:log_section]
2025-06-09 20:59:47,817 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:114:log_section]
2025-06-09 20:59:47,817 - INFO - 🧠 Initializing custom L2R model for outputting per-token relevance scores per ICD-10 codes. - [learning2rank.py:4047:main]
2025-06-09 20:59:49,367 - INFO - 🔧 Registering LTRModel with transformers AutoModel 🚀 - [learning2rank.py:1899:define_model]
2025-06-09 20:59:49,367 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:110:log_section]
2025-06-09 20:59:49,367 - INFO - + TRAINING PREPARATION + - [learning2rank.py:111:log_section]
2025-06-09 20:59:49,367 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:114:log_section]
2025-06-09 20:59:49,367 - INFO - ⚙️ Preparing training components and optimizers... - [learning2rank.py:4054:main]
2025-06-09 20:59:49,399 - INFO - 🖥️ Device: NVIDIA GH200 480GB - [learning2rank.py:2061:log_training_configuration]
2025-06-09 20:59:49,400 - INFO - 🔋 CUDA Available: True - [learning2rank.py:2064:log_training_configuration]
2025-06-09 20:59:49,400 - INFO - 💾 CUDA Device Count: 1 - [learning2rank.py:2065:log_training_configuration]
2025-06-09 20:59:49,400 - INFO -
📋 Training Configuration 📋
+----------+-----------------------------+--------------------------------------------------+
| 🌟 Emoji | 🏷️ Parameter | 📊 Value |
+----------+-----------------------------+--------------------------------------------------+
| 📁 | Output Directory | ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b |
| 🔁 | Training Epochs | 4 |
| 🏋️ | Train Batch Size | 1 |
| 🔍 | Eval Batch Size | 1 |
| 📊 | Gradient Accumulation Steps | 4 |
| 🚀 | Learning Rate | 0.0001 |
| 🌅 | Warmup Steps | 0 |
| 💾 | Save Strategy | epoch |
| 💾 | Save Total Limit | 10 |
| 📊 | Evaluation Strategy | epoch |
| 🎯 | Best Model Metric | ndcg@25 |
| 📝 | Logging Strategy | steps (every 10 steps) |
| 🌐 | Push to Hub | True |
| 🌐 | Hub Model ID | deb101/mistral-7b-instruct-v0.3-mimic4-adapt-l2r |
| 🔢 | Steps per Epoch | 43 |
| 🔢 | Total Training Steps | 172 |
| 🔢 | Evaluation Steps | 35 |
| 📊 | Training Dataset Size | 173 samples 🏋️ |
| 📊 | Evaluation Dataset Size | 35 samples 🔍 |
+----------+-----------------------------+--------------------------------------------------+ - [learning2rank.py:2053:log_training_args]
2025-06-09 20:59:49,401 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:110:log_section]
2025-06-09 20:59:49,401 - INFO - + MODEL TRAINING + - [learning2rank.py:111:log_section]
2025-06-09 20:59:49,401 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:114:log_section]
2025-06-09 20:59:49,401 - INFO - 🏋️ Starting model training process... - [learning2rank.py:4074:main]
2025-06-09 20:59:49,401 - INFO - 🏁 Preparing Custom Trainer 🛠️ - [learning2rank.py:3254:train_model]
2025-06-09 20:59:49,440 - INFO - We are registering the tokenizer mistralai/Mistral-7B-Instruct-v0.3 in Custom Trainer - [learning2rank.py:2693:__init__]
2025-06-09 20:59:49,440 - INFO - 🏋️ Commencing Model Training 💪 - [learning2rank.py:3295:train_model]
2025-06-09 20:59:49,717 - INFO - 🚀 Starting Training... - [learning2rank.py:2443:on_train_begin]
2025-06-09 21:00:09,291 - INFO -
🚂 Training Metrics (Step 10) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -4.64769e+16 |
+---------------+--------------+
| grad_norm | nan |
+---------------+--------------+
| learning_rate | 0.0001 |
+---------------+--------------+
| epoch | 0.231214 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:00:23,861 - INFO -
🚂 Training Metrics (Step 20) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -4.10389e+17 |
+---------------+--------------+
| grad_norm | nan |
+---------------+--------------+
| learning_rate | 0.0001 |
+---------------+--------------+
| epoch | 0.462428 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:00:38,479 - INFO -
🚂 Training Metrics (Step 30) 🚂
+---------------+------------+
| Metric | Value |
+===============+============+
| loss | -6.489e+16 |
+---------------+------------+
| grad_norm | nan |
+---------------+------------+
| learning_rate | 0.0001 |
+---------------+------------+
| epoch | 0.693642 |
+---------------+------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:00:53,093 - INFO -
🚂 Training Metrics (Step 40) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -1.34976e+18 |
+---------------+--------------+
| grad_norm | nan |
+---------------+--------------+
| learning_rate | 0.0001 |
+---------------+--------------+
| epoch | 0.924855 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:00:57,833 - INFO - Removing 'token_type_ids' from eval_dataset as they are not needed. - [learning2rank.py:2809:evaluate]
2025-06-09 21:04:13,175 - INFO -
🔍 Evaluation Metrics 🔍
+-------------------+--------------+
| Metric | Value |
+===================+==============+
| eval_loss | -1.00105e+17 |
+-------------------+--------------+
| eval_ndcg | 0.95541 |
+-------------------+--------------+
| eval_ndcg@25 | 0.216384 |
+-------------------+--------------+
| eval_precision@25 | 0.04 |
+-------------------+--------------+ - [learning2rank.py:2380:on_evaluate]
2025-06-09 21:04:14,390 - INFO - 💾 Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-44 - [learning2rank.py:2933:_save]
2025-06-09 21:04:14,418 - INFO - ⚙️ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-44 - [learning2rank.py:2938:_save]
2025-06-09 21:04:14,419 - INFO - 📋 Saved files in ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-44:
+---------+-------------------+------------+
| Index | Saved File | Size |
+=========+===================+============+
| 1 | training_args.bin | 0.01 MB |
+---------+-------------------+------------+
| 2 | model.safetensors | 4122.74 MB |
+---------+-------------------+------------+
| 3 | config.json | 0.38 MB |
+---------+-------------------+------------+ - [learning2rank.py:2955:_save]
2025-06-09 21:04:27,975 - INFO -
🚂 Training Metrics (Step 50) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -3.53393e+18 |
+---------------+--------------+
| grad_norm | nan |
+---------------+--------------+
| learning_rate | 0.0001 |
+---------------+--------------+
| epoch | 1.13873 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:04:42,553 - INFO -
🚂 Training Metrics (Step 60) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -1.14824e+17 |
+---------------+--------------+
| grad_norm | nan |
+---------------+--------------+
| learning_rate | 9.9e-05 |
+---------------+--------------+
| epoch | 1.36994 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:04:57,080 - INFO -
🚂 Training Metrics (Step 70) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -6.82879e+18 |
+---------------+--------------+
| grad_norm | 2.36978e+18 |
+---------------+--------------+
| learning_rate | 9.4e-05 |
+---------------+--------------+
| epoch | 1.60116 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:05:11,630 - INFO -
🚂 Training Metrics (Step 80) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -8.96262e+16 |
+---------------+--------------+
| grad_norm | 1.67389e+18 |
+---------------+--------------+
| learning_rate | 8.8e-05 |
+---------------+--------------+
| epoch | 1.83237 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:05:22,189 - INFO - Removing 'token_type_ids' from eval_dataset as they are not needed. - [learning2rank.py:2809:evaluate]
2025-06-09 21:08:35,777 - INFO -
🔍 Evaluation Metrics 🔍
+-------------------+------------+
| Metric | Value |
+===================+============+
| eval_loss | -2.959e+17 |
+-------------------+------------+
| eval_ndcg | 0.956161 |
+-------------------+------------+
| eval_ndcg@25 | 0.6676 |
+-------------------+------------+
| eval_precision@25 | 0.776 |
+-------------------+------------+ - [learning2rank.py:2380:on_evaluate]
2025-06-09 21:08:36,992 - INFO - 💾 Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-88 - [learning2rank.py:2933:_save]
2025-06-09 21:08:37,019 - INFO - ⚙️ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-88 - [learning2rank.py:2938:_save]
2025-06-09 21:08:37,020 - INFO - 📋 Saved files in ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-88:
+---------+-------------------+------------+
| Index | Saved File | Size |
+=========+===================+============+
| 1 | training_args.bin | 0.01 MB |
+---------+-------------------+------------+
| 2 | model.safetensors | 4122.74 MB |
+---------+-------------------+------------+
| 3 | config.json | 0.38 MB |
+---------+-------------------+------------+ - [learning2rank.py:2955:_save]
2025-06-09 21:08:44,917 - INFO -
🚂 Training Metrics (Step 90) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -6.79621e+17 |
+---------------+--------------+
| grad_norm | 2.89885e+17 |
+---------------+--------------+
| learning_rate | 8.3e-05 |
+---------------+--------------+
| epoch | 2.04624 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:08:59,431 - INFO -
🚂 Training Metrics (Step 100) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -2.13981e+17 |
+---------------+--------------+
| grad_norm | 5.44041e+17 |
+---------------+--------------+
| learning_rate | 7.7e-05 |
+---------------+--------------+
| epoch | 2.27746 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:09:13,894 - INFO -
🚂 Training Metrics (Step 110) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -4.69121e+18 |
+---------------+--------------+
| grad_norm | nan |
+---------------+--------------+
| learning_rate | 7.3e-05 |
+---------------+--------------+
| epoch | 2.50867 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:09:28,310 - INFO -
🚂 Training Metrics (Step 120) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -3.82384e+17 |
+---------------+--------------+
| grad_norm | 9.07643e+17 |
+---------------+--------------+
| learning_rate | 6.7e-05 |
+---------------+--------------+
| epoch | 2.73988 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:09:42,759 - INFO -
🚂 Training Metrics (Step 130) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -1.04216e+18 |
+---------------+--------------+
| grad_norm | 2.78575e+17 |
+---------------+--------------+
| learning_rate | 6.1e-05 |
+---------------+--------------+
| epoch | 2.9711 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:09:44,592 - INFO - Removing 'token_type_ids' from eval_dataset as they are not needed. - [learning2rank.py:2809:evaluate]
2025-06-09 21:12:57,810 - INFO -
🔍 Evaluation Metrics 🔍
+-------------------+--------------+
| Metric | Value |
+===================+==============+
| eval_loss | -4.17271e+17 |
+-------------------+--------------+
| eval_ndcg | 0.956751 |
+-------------------+--------------+
| eval_ndcg@25 | 0.776161 |
+-------------------+--------------+
| eval_precision@25 | 0.926857 |
+-------------------+--------------+ - [learning2rank.py:2380:on_evaluate]
2025-06-09 21:12:59,006 - INFO - 💾 Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-132 - [learning2rank.py:2933:_save]
2025-06-09 21:12:59,033 - INFO - ⚙️ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-132 - [learning2rank.py:2938:_save]
2025-06-09 21:12:59,034 - INFO - 📋 Saved files in ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-132:
+---------+-------------------+------------+
| Index | Saved File | Size |
+=========+===================+============+
| 1 | training_args.bin | 0.01 MB |
+---------+-------------------+------------+
| 2 | model.safetensors | 4122.74 MB |
+---------+-------------------+------------+
| 3 | config.json | 0.38 MB |
+---------+-------------------+------------+ - [learning2rank.py:2955:_save]
2025-06-09 21:13:15,686 - INFO -
🚂 Training Metrics (Step 140) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -1.11827e+19 |
+---------------+--------------+
| grad_norm | 1.68992e+17 |
+---------------+--------------+
| learning_rate | 5.6e-05 |
+---------------+--------------+
| epoch | 3.18497 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:13:30,085 - INFO -
🚂 Training Metrics (Step 150) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -6.84765e+18 |
+---------------+--------------+
| grad_norm | 4.36442e+17 |
+---------------+--------------+
| learning_rate | 5.1e-05 |
+---------------+--------------+
| epoch | 3.41619 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:13:44,519 - INFO -
🚂 Training Metrics (Step 160) 🚂
+---------------+-------------+
| Metric | Value |
+===============+=============+
| loss | -4.6995e+17 |
+---------------+-------------+
| grad_norm | 2.44171e+17 |
+---------------+-------------+
| learning_rate | 4.5e-05 |
+---------------+-------------+
| epoch | 3.6474 |
+---------------+-------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:13:58,970 - INFO -
🚂 Training Metrics (Step 170) 🚂
+---------------+--------------+
| Metric | Value |
+===============+==============+
| loss | -3.47634e+17 |
+---------------+--------------+
| grad_norm | 3.07701e+17 |
+---------------+--------------+
| learning_rate | 4e-05 |
+---------------+--------------+
| epoch | 3.87861 |
+---------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:14:03,062 - INFO - 💾 Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-172 - [learning2rank.py:2933:_save]
2025-06-09 21:14:03,090 - INFO - ⚙️ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-172 - [learning2rank.py:2938:_save]
2025-06-09 21:14:03,091 - INFO - 📋 Saved files in ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-172:
+---------+-------------------+------------+
| Index | Saved File | Size |
+=========+===================+============+
| 1 | training_args.bin | 0.01 MB |
+---------+-------------------+------------+
| 2 | model.safetensors | 4122.74 MB |
+---------+-------------------+------------+
| 3 | config.json | 0.38 MB |
+---------+-------------------+------------+ - [learning2rank.py:2955:_save]
2025-06-09 21:14:03,312 - INFO - Removing 'token_type_ids' from eval_dataset as they are not needed. - [learning2rank.py:2809:evaluate]
2025-06-09 21:17:17,722 - INFO -
🔍 Evaluation Metrics 🔍
+-------------------+--------------+
| Metric | Value |
+===================+==============+
| eval_loss | -4.17112e+17 |
+-------------------+--------------+
| eval_ndcg | 0.956852 |
+-------------------+--------------+
| eval_ndcg@25 | 0.808041 |
+-------------------+--------------+
| eval_precision@25 | 0.914286 |
+-------------------+--------------+ - [learning2rank.py:2380:on_evaluate]
2025-06-09 21:17:21,247 - INFO - 💾 Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-172 - [learning2rank.py:2933:_save]
2025-06-09 21:17:21,275 - INFO - ⚙️ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-172 - [learning2rank.py:2938:_save]
2025-06-09 21:17:21,277 - INFO - 📋 Saved files in ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-172:
+---------+--------------------+------------+
| Index | Saved File | Size |
+=========+====================+============+
| 1 | training_args.bin | 0.01 MB |
+---------+--------------------+------------+
| 2 | optimizer.pt | 352.39 MB |
+---------+--------------------+------------+
| 3 | model.safetensors | 4122.74 MB |
+---------+--------------------+------------+
| 4 | scaler.pt | 0.00 MB |
+---------+--------------------+------------+
| 5 | config.json | 0.38 MB |
+---------+--------------------+------------+
| 6 | scheduler.pt | 0.00 MB |
+---------+--------------------+------------+
| 7 | trainer_state.json | 0.00 MB |
+---------+--------------------+------------+
| 8 | rng_state.pth | 0.01 MB |
+---------+--------------------+------------+ - [learning2rank.py:2955:_save]
2025-06-09 21:17:21,593 - INFO - 📂 Loading best model from ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-172 - [learning2rank.py:3011:_load_best_model]
2025-06-09 21:17:21,593 - INFO - 🖥️ Model is on device: cuda:0 - [learning2rank.py:3021:_load_best_model]
2025-06-09 21:17:21,658 - INFO - 🔑 Key order comparison:
+---------+--------------------------------------------------------------------------+-------------------------------------------------------------------------------------------+
| Index | Saved state_dict Keys | Model state_dict Keys |
+=========+==========================================================================+===========================================================================================+
| 1 | ground_model.base_model.model.lm_head.weight | label_embeddings.weight |
+---------+--------------------------------------------------------------------------+-------------------------------------------------------------------------------------------+
| 2 | ground_model.base_model.model.model.embed_tokens.weight | ground_model.base_model.model.model.embed_tokens.weight |
+---------+--------------------------------------------------------------------------+-------------------------------------------------------------------------------------------+
| 3 | ground_model.base_model.model.model.layers.0.input_layernorm.weight | ground_model.base_model.model.model.layers.0.self_attn.q_proj.base_layer.weight |
+---------+--------------------------------------------------------------------------+-------------------------------------------------------------------------------------------+
| 4 | ground_model.base_model.model.model.layers.0.mlp.down_proj.weight | ground_model.base_model.model.model.layers.0.self_attn.q_proj.base_layer.weight.absmax |
+---------+--------------------------------------------------------------------------+-------------------------------------------------------------------------------------------+
| 5 | ground_model.base_model.model.model.layers.0.mlp.down_proj.weight.absmax | ground_model.base_model.model.model.layers.0.self_attn.q_proj.base_layer.weight.quant_map |
+---------+--------------------------------------------------------------------------+-------------------------------------------------------------------------------------------+ - [learning2rank.py:3045:_load_best_model]
2025-06-09 21:17:22,481 - INFO - Loaded best model weights from ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/checkpoint-172/model.safetensors - [learning2rank.py:3062:_load_best_model]
2025-06-09 21:17:22,520 - INFO - ✔️ Weight for label_embeddings.weight matches between saved and loaded state_dict - [learning2rank.py:3074:_load_best_model]
2025-06-09 21:17:22,576 - INFO - ✔️ Weight for ground_model.base_model.model.model.embed_tokens.weight matches between saved and loaded state_dict - [learning2rank.py:3074:_load_best_model]
2025-06-09 21:17:22,593 - INFO -
🚂 Training Metrics (Step 172) 🚂
+--------------------------+--------------+
| Metric | Value |
+==========================+==============+
| train_runtime | 1052.88 |
+--------------------------+--------------+
| train_samples_per_second | 0.657 |
+--------------------------+--------------+
| train_steps_per_second | 0.163 |
+--------------------------+--------------+
| total_flos | 7.58799e+15 |
+--------------------------+--------------+
| train_loss | -2.22959e+18 |
+--------------------------+--------------+
| epoch | 3.92485 |
+--------------------------+--------------+ - [learning2rank.py:2361:on_log]
2025-06-09 21:17:22,594 - INFO - Training Completed! - [learning2rank.py:2510:on_train_end]
2025-06-09 21:17:22,669 - INFO - 📊 Training loss plot saved as '../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/train_loss_plot.png' - [learning2rank.py:2613:on_train_end]
2025-06-09 21:17:22,731 - INFO - 📊 Evaluation loss plot saved as '../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/eval_loss_plot.png' - [learning2rank.py:2627:on_train_end]
2025-06-09 21:17:22,791 - INFO - 📊 Evaluation metric plot saved as '../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/eval_ndcg@25_plot.png' - [learning2rank.py:2648:on_train_end]
2025-06-09 21:17:22,792 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:110:log_section]
2025-06-09 21:17:22,792 - INFO - + MODEL SAVING + - [learning2rank.py:111:log_section]
2025-06-09 21:17:22,792 - INFO - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ - [learning2rank.py:114:log_section]
2025-06-09 21:17:22,792 - INFO - 💾 Saving trained model and pushing to Hugging Face Hub... - [learning2rank.py:4089:main]
2025-06-09 21:21:37,872 - INFO - 📁 Creating/using output directory: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b - [learning2rank.py:3329:save_and_push]
2025-06-09 21:21:39,319 - INFO - 💾 Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b - [learning2rank.py:2933:_save]
2025-06-09 21:21:41,035 - INFO - ⚙️ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b - [learning2rank.py:2938:_save]
2025-06-09 21:21:41,049 - INFO - 📋 Saved files in ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b:
+---------+-----------------------+------------+
| Index | Saved File | Size |
+=========+=======================+============+
| 1 | eval_loss_plot.png | 0.03 MB |
+---------+-----------------------+------------+
| 2 | training_args.bin | 0.01 MB |
+---------+-----------------------+------------+
| 3 | model.safetensors | 4122.74 MB |
+---------+-----------------------+------------+
| 4 | eval_ndcg@25_plot.png | 0.03 MB |
+---------+-----------------------+------------+
| 5 | config.json | 0.38 MB |
+---------+-----------------------+------------+
| 6 | train_loss_plot.png | 0.04 MB |
+---------+-----------------------+------------+ - [learning2rank.py:2955:_save]
2025-06-09 21:21:44,636 - INFO - 💾 Model weights saved in safetensors format: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b - [learning2rank.py:2933:_save]
2025-06-09 21:21:46,152 - INFO - ⚙️ Config saved in checkpoint: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b - [learning2rank.py:2938:_save]
2025-06-09 21:21:46,165 - INFO - 📋 Saved files in ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b:
+---------+-----------------------+------------+
| Index | Saved File | Size |
+=========+=======================+============+
| 1 | eval_loss_plot.png | 0.03 MB |
+---------+-----------------------+------------+
| 2 | training_args.bin | 0.01 MB |
+---------+-----------------------+------------+
| 3 | model.safetensors | 4122.74 MB |
+---------+-----------------------+------------+
| 4 | eval_ndcg@25_plot.png | 0.03 MB |
+---------+-----------------------+------------+
| 5 | config.json | 0.38 MB |
+---------+-----------------------+------------+
| 6 | train_loss_plot.png | 0.04 MB |
+---------+-----------------------+------------+ - [learning2rank.py:2955:_save]
2025-06-09 21:23:01,078 - INFO - 💾 Model saved to: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b - [learning2rank.py:3333:save_and_push]
2025-06-09 21:23:01,363 - INFO - 🖌️ Tokenizer saved to: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b - [learning2rank.py:3337:save_and_push]
2025-06-09 21:23:01,837 - INFO - 📊 Lookup table saved to: ../tmp/MIMIC4_DEMO/mimic4_l2rtrain_mistral7b/lookup_table.pt - [learning2rank.py:3342:save_and_push]