Update app.py

app.py

@@ -16,7 +16,7 @@ from datetime import datetime
 
 #model_config = getconfig("model_params.cfg")
 ###########
-# ToDo move to functions
+# ToDo move functions to utils and model specifications to config file!
 # Configuration for the dedicated model
 DEDICATED_MODEL = "meta-llama/Llama-3.1-8B-Instruct"
 DEDICATED_ENDPOINT = "https://qu2d8m6dmsollhly.us-east-1.aws.endpoints.huggingface.cloud"
@@ -29,7 +29,8 @@ def get_rag_answer(query, top_results):
     if len(context) > max_context_chars:
         context = context[:max_context_chars]
     prompt = (
-        "Using the following context, answer the question concisely. "
+        "You are a project portfolio adviser at the development cooperation GIZ. "
+        "Using the following context, answer the question precisely. "
         "Only output the final answer below, without repeating the context or question.\n\n"
         f"Context:\n{context}\n\n"
         f"Question: {query}\n\n"
@@ -38,7 +39,7 @@ def get_rag_answer(query, top_results):
     headers = {"Authorization": f"Bearer {WRITE_ACCESS_TOKEN}"}
     payload = {
         "inputs": prompt,
-        "parameters": {"max_new_tokens": 150}
+        "parameters": {"max_new_tokens": 200}
     }
     response = requests.post(DEDICATED_ENDPOINT, headers=headers, json=payload)
     if response.status_code == 200:
@@ -78,7 +79,12 @@ def get_crs_options(_client, collection_name):
         crs_key = metadata.get("crs_key", "").strip()
         crs_value = metadata.get("crs_value", "").strip()
         if crs_key or crs_value:
-            crs_combined = f"{crs_key}: {crs_value}"
+            # Convert crs_value to integer if possible:
+            try:
+                crs_int = int(float(crs_value))
+            except:
+                crs_int = crs_value
+            crs_combined = f"{crs_key}: {crs_int}"
             crs_set.add(crs_combined)
     return sorted(crs_set)