🌘w🌖

app.py

@@ -21,7 +21,7 @@ tokenizer = AutoTokenizer.from_pretrained("google/gemma-7b-it", token=token)
 device = torch.device("cuda")
 model = model.to(device)
 RAG = SentenceTransformer("mixedbread-ai/mxbai-embed-large-v1")
-TOP_K = 3
+TOP_K = 2
 HEADER = "\n# RESOURCES:\n"
 # prepare data
 # since data is too big we will only select the first 3K lines
@@ -50,7 +50,7 @@ def prepare_prompt(query, retrieved_examples):
     urls = retrieved_examples["url"][::-1]
     titles = titles[::-1]
     for i in range(TOP_K):
-        prompt += f"Title: {titles[i]}, Text: {texts[i]}\n"
+        prompt += f"* {texts[i]}\n"
     return prompt, zip(titles, urls)