Fix bad inference

backend/runner/inference.py

@@ -72,12 +72,14 @@ def load_embeddings_from_hf():
     try:
         print(f" Loading embeddings from {ARTEFACT_EMBEDDINGS_DATASET}...")
         
-        if not EMBEDDINGS_DATASETS:
+        # Call the function to get the actual dictionary
+        embeddings_datasets = EMBEDDINGS_DATASETS()
+        if not embeddings_datasets:
             print("❌ No embeddings datasets loaded")
             return None
             
         # Check if we're using direct download
-        if EMBEDDINGS_DATASETS.get('use_direct_download', False):
+        if embeddings_datasets.get('use_direct_download', False):
             print("✅ Using direct file download for embeddings")
             
             # Download the safetensors files
@@ -598,11 +600,13 @@ def st_load_file(file_path: Path) -> Any:
 def load_embedding_for_sentence(sentence_id: str, model_type: str = "clip") -> Optional[torch.Tensor]:
     """Load a single embedding for a specific sentence using streaming"""
     try:
-        if not EMBEDDINGS_DATASETS or not EMBEDDINGS_DATASETS.get('use_streaming', False):
+        # Call the function to get the actual dictionary
+        embeddings_datasets = EMBEDDINGS_DATASETS()
+        if not embeddings_datasets or not embeddings_datasets.get('use_streaming', False):
             print("❌ Streaming embeddings not available")
             return None
             
-        dataset = EMBEDDINGS_DATASETS['streaming_dataset']
+        dataset = embeddings_datasets['streaming_dataset']
         
         # Search for the sentence in the streaming dataset
         for item in dataset:
@@ -626,11 +630,13 @@ def load_embedding_for_sentence(sentence_id: str, model_type: str = "clip") -> O
 def get_top_k_embeddings(query_embedding: torch.Tensor, k: int = 10, model_type: str = "clip") -> List[Tuple[str, float]]:
     """Get top-k most similar embeddings using streaming"""
     try:
-        if not EMBEDDINGS_DATASETS or not EMBEDDINGS_DATASETS.get('use_streaming', False):
+        # Call the function to get the actual dictionary
+        embeddings_datasets = EMBEDDINGS_DATASETS()
+        if not embeddings_datasets or not embeddings_datasets.get('use_streaming', False):
             print("❌ Streaming embeddings not available")
             return []
             
-        dataset = EMBEDDINGS_DATASETS['streaming_dataset']
+        dataset = embeddings_datasets['streaming_dataset']
         similarities = []
         
         # Process embeddings in batches to avoid memory issues
@@ -720,10 +726,12 @@ def run_inference_streaming(
         print(f"✅ Image embedding computed successfully")
 
         # Get streaming dataset
-        if not EMBEDDINGS_DATASETS or not EMBEDDINGS_DATASETS.get('use_streaming', False):
+        # Call the function to get the actual dictionary
+        embeddings_datasets = EMBEDDINGS_DATASETS()
+        if not embeddings_datasets or not embeddings_datasets.get('use_streaming', False):
             raise ValueError("Streaming embeddings not available")
             
-        dataset = EMBEDDINGS_DATASETS['streaming_dataset']
+        dataset = embeddings_datasets['streaming_dataset']
         
         # Process embeddings in streaming mode
         results = []