data/datasets/三国演义/splits/split_summary.json

@@ -1,24 +0,0 @@
-{
-  "input_dir": "data/datasets/三国演义",
-  "output_dir": "data/datasets/三国演义/splits",
-  "split_ratios": {
-    "train": 0.8,
-    "validation": 0.1,
-    "test": 0.1
-  },
-  "seed": 42,
-  "datasets": {
-    "bge_m3": {
-      "total_samples": 31370,
-      "train_samples": 25096,
-      "val_samples": 3137,
-      "test_samples": 3137
-    },
-    "reranker": {
-      "total_samples": 26238,
-      "train_samples": 20990,
-      "val_samples": 2623,
-      "test_samples": 2625
-    }
-  }
-}

tests/test_installation.py

@@ -525,12 +525,7 @@ class ComprehensiveTestSuite:
 
                 # Test saving
                 state_dict = {
-                    'model_state_dict': {
+                    'model_state_dict': torch.randn(10, 10),
-                        'layer1.weight': torch.randn(10, 5),
-                        'layer1.bias': torch.randn(10),
-                        'layer2.weight': torch.randn(3, 10),
-                        'layer2.bias': torch.randn(3)
-                    },
                     'optimizer_state_dict': {'lr': 0.001},
                     'step': 100
                 }

utils/checkpoint.py

@@ -86,7 +86,7 @@ class CheckpointManager:
                 model_path = os.path.join(temp_dir, 'pytorch_model.bin')
                 # Add extra safety check for model state dict
                 model_state = state_dict['model_state_dict']
-                if model_state is None or (isinstance(model_state, dict) and len(model_state) == 0):
+                if not model_state:
                     raise RuntimeError("Model state dict is empty")
                 
                 self._safe_torch_save(model_state, model_path)