minor changes

Author: paramkpr

.DS_Store

@@ -76,4 +76,9 @@ outputs/
 *.DS_Store
 .DS_Store
 .DS_Store
-.DS_Store?
+.DS_Store?
+
+.venv/
+.env/
+
+.DS_Store

.gitignore

@@ -54,8 +54,10 @@ To run on `weftdrive`:
    1. `export WANDB_API_KEY="..."`
    2. `python -m wandb login`
 7. Create the logs directory and file: `mkdir -p ~/scratch/paramkapur/logs` and `touch ~/scratch/paramkapur/logs/$(date +%Y%m%d_%H%M).log`
-8. Run the training script: `nohup /srv/gpurun.pl python src/cli/01_train_teacher.py configs/teacher/stt2_hf.yaml > ~/scratch/paramkapur/logs/$(date +%Y%m%d_%H%M).log 2>&1 &`
+8. Run the training script: `nohup /srv/gpurun.pl python src/cli/01_train_teacher.py configs/teacher/sst2_hf.yaml > ~/scratch/paramkapur/logs/$(date +%Y%m%d_%H%M).log 2>&1 &`
 
 
 /scratch/paramkapur/data/clean/clean
 
+
+nohup /srv/gpurun.pl python src/cli/01_train_teacher.py configs/teacher/sst2_hf.yaml > ~/scratch/paramkapur/logs/$(date +%Y%m%d_%H%M).log 2>&1 &

README.md

@@ -4,7 +4,7 @@ model:
   use_fast_tokenizer: true
 
 data:
-  dataset_path: "~/scratch/paramkapur/data/clean/clean" # Use HF dataset identifier
+  dataset_path: "./data/clean/" # Use HF dataset identifier
   max_len: 32
   train_split: "train"
   validation_split: "val"

configs/teacher/sst2_hf.yaml

@@ -1,22 +1,22 @@
-torch>=1.9.0
+torch
 transformers
-datasets>=1.11.0
-numpy>=1.19.5
-scikit-learn>=0.24.2
-pandas>=1.3.0
-matplotlib>=3.4.2
-seaborn>=0.11.1
-tqdm>=4.61.2
-datasets>=3.5.0
+datasets
+numpy
+scikit-learn
+pandas
+matplotlib
+seaborn
+tqdm
+datasets
 
-accelerate==1.6.0
+accelerate
 
 # Configuration and Metrics
-PyYAML>=6.0.2
-scikit-learn>=1.6.1
+PyYAML
+scikit-learn
 
 # Logging
-wandb>=0.19.9
+wandb
 Cmake
 sentencepiece
 protobuf

requirements.txt

@@ -8,9 +8,20 @@ def compute_metrics(p):
     labels = p.label_ids
     precision, recall, f1, _ = precision_recall_fscore_support(labels, preds, average='binary') # Assuming binary
     acc = accuracy_score(labels, preds)
+    
+    # Calculate confusion matrix metrics
+    true_positives = np.sum((preds == 1) & (labels == 1))
+    false_positives = np.sum((preds == 1) & (labels == 0))
+    true_negatives = np.sum((preds == 0) & (labels == 0))
+    false_negatives = np.sum((preds == 0) & (labels == 1))
+    
     return {
         'accuracy': acc,
         'f1': f1,
         'precision': precision,
-        'recall': recall
+        'recall': recall,
+        'true_positives': int(true_positives),
+        'false_positives': int(false_positives), 
+        'true_negatives': int(true_negatives),
+        'false_negatives': int(false_negatives)
     }