Revised init commit

tests/test_full_training.py

@@ -56,19 +56,22 @@ def main():
         model_name_or_path="./models/bge-m3",
         train_data_path=train_file,
         output_dir="output/test-lr-scheduler",
-        num_train_epochs=2,
-        per_device_train_batch_size=2,
-        gradient_accumulation_steps=1,  # Small for testing
-        learning_rate=1e-4,  # Higher for testing
+        num_train_epochs=3,
+        per_device_train_batch_size=1,  # Smaller to ensure more steps
+        gradient_accumulation_steps=2,  # More accumulation steps for stability
+        learning_rate=1e-5,  # Lower learning rate for stability
         logging_steps=1,
-        save_steps=10,
-        fp16=False,
+        save_steps=0,  # Disable frequent checkpointing during test
+        eval_steps=0,  # Disable evaluation during test
+        fp16=False,  # Disable mixed precision for stability
+        bf16=False,  # Disable bfloat16
         use_amp=False,
         dataloader_num_workers=0,
         overwrite_output_dir=True,
         use_self_distill=False,
         use_hard_negatives=False,
-        temperature=0.1  # Fixed temperature
+        temperature=0.1,  # Fixed temperature
+        save_total_limit=1  # Limit checkpoint storage
     )
     
     try: