learning how to code with the post-introspector

app.py

@@ -65,18 +65,16 @@ class EmbeddingGenerator:
                 {"role": "user", "content": escaped_input_text}  
             ]  
         )  
-        intention_output = intention_completion.choices[0].message['content']  
-  
+        intention_output = intention_completion.choices[0].message.content  
         # Parse and route the intention  
         parsed_task = parse_and_route(intention_output)  
-        selected_task = list(parsed_task.keys())[0]  
-  
+        selected_task = parsed_task  
         # Construct the prompt  
-        try:  
+        if selected_task in tasks:
             task_description = tasks[selected_task]  
-        except KeyError:  
+        else:
+            task_description = tasks["DEFAULT"]
             print(f"Selected task not found: {selected_task}")  
-            return f"Error: Task '{selected_task}' not found. Please select a valid task."  
   
         query_prefix = f"Instruct: {task_description}\nQuery: "  
         queries = [escaped_input_text]  
@@ -89,13 +87,14 @@ class EmbeddingGenerator:
                 {"role": "user", "content": escaped_input_text}  
             ]  
         )  
-        metadata_output = metadata_completion.choices[0].message['content']  
+        metadata_output = metadata_completion.choices[0].message.content
         metadata = self.extract_metadata(metadata_output)  
   
         # Get the embeddings  
         with torch.no_grad():  
             inputs = self.tokenizer(queries, return_tensors='pt', padding=True, truncation=True, max_length=4096).to(self.device)  
-            outputs = self.model(**inputs)  
+            outputs = self.model(**inputs)
+            query_embeddings = outputs["sentence_embeddings"].mean(dim=1)    
             query_embeddings = outputs.last_hidden_state.mean(dim=1)  
   
             # Normalize embeddings  
@@ -118,7 +117,7 @@ class MyEmbeddingFunction(EmbeddingFunction):
         self.embedding_generator = embedding_generator  
   
     def __call__(self, input: Documents) -> (Embeddings, list):  
-        embeddings_with_metadata = [self.embedding_generator.compute_embeddings(doc) for doc in input]  
+        embeddings_with_metadata = [self.embedding_generator.compute_embeddings(doc.page_content) for doc in input]  
         embeddings = [item[0] for item in embeddings_with_metadata]  
         metadata = [item[1] for item in embeddings_with_metadata]  
         embeddings_flattened = [emb for sublist in embeddings for emb in sublist]