hold up let it cook

app.py

@@ -12,29 +12,32 @@ with open("uni_dataset_embedding_friendly.txt", "r", encoding="utf-8") as file:
 # Print the text below
 print("success")
 
-chunks = [chunk.strip() for chunk in uni_dataset_text.split("\n\n") if chunk.strip()]
+chunks = [chunk.strip() for chunk in uni_dataset_text.split("\n---\n") if chunk.strip()]
 embedder = SentenceTransformer('all-MiniLM-L6-v2')
 chunk_embeddings = embedder.encode(chunks, convert_to_tensor= True)
+norm_chunk_embeddings = torch.nn.functional.normalize(chunk_embeddings, dim=1)
 
 def get_relevant_context(query, top_k=3):
-    query_embedding = embedder.encode(query, convert_to_tensor = True)
-    query_embedding = query_embedding / query_embedding.norm()
-    norm_chunk_embeddings = chunk_embeddings / chunk_embeddings.norm(dim=1, keepdim=True)
+    query_embedding = embedder.encode(query, convert_to_tensor=True)
+    query_embedding = torch.nn.functional.normalize(query_embedding, dim=0)
     similarities = torch.matmul(norm_chunk_embeddings, query_embedding)
-    top_k_indices = torch.topk(similarities, k=top_k).indices.cpu().numpy()
-    context = "\n\n".join([chunks[i] for i in top_k_indices])
-    return context
+    k = min(top_k, similarities.shape[0])
+    top_k_indices = torch.topk(similarities, k=k).indices.cpu().tolist()
+    selected = [chunks[i] for i in top_k_indices]
+    return selected, top_k_indices
 
     
 client = InferenceClient("microsoft/phi-4")
 def respond(message, history):
-    messages = [{"role": "system", "content": "you are a realistic and friendly career advisor to help secondary school students with important decisions such as the university courses they should apply to, careers to pursue, etc. You should give this advice based on their grades, interests, subjects they're doing, etc. Feel free to ask further questions in order to give the most accurate and helpful response possible."}]
+    messages = [{"role": "system", "content": "you are a realistic and friendly career advisor to help secondary school students with important decisions such as the university courses they should apply to, careers to pursue, etc. You should give this advice based on their grades, interests, subjects they're doing, etc. Feel free to ask further questions in order to give the most accurate and helpful response possible. Prioritise answering questions with the information supplied. Be concise, helpful and, if useful, ask brief follow-up questions."}]
     if history:
         messages.extend(history)
     messages.append({"role": "user", "content":message})
     response = client.chat_completion(
-        messages,
-        max_tokens=500
+        messages = messages,
+        temperature = 0.0,
+        max_tokens=500,
+        top_p = 1.0
     )
     return response['choices'][0]['message']['content'].strip()