feat: use microsoft phi-2

app.py

@@ -16,7 +16,7 @@ with open("vector_db/metadata.pkl", "rb") as f:
 
 ST = SentenceTransformer("BAAI/bge-large-en-v1.5")
 
-model_id = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
+model_id = "microsoft/phi-2"
 bnb = BitsAndBytesConfig(
     load_in_4bit=True,
     bnb_4bit_use_double_quant=True,
@@ -58,7 +58,7 @@ def retrieve(q, k=3):
 
 def make_prompt(q, docs):
     context = "\n\n".join(f"Title: {doc['title']}\nPages: {doc['pages']}" for doc in docs)
-    return f"{SYS}\n\nContext:\n{context}\n\nQuestion:\n{q}\n\nAnswer:"
+    return f"Instruct:{SYS}\n\nContext:\n{context}\n\nQuestion:\n{q}\n\nOutput:"
 
 @spaces.GPU()
 def qa_fn(question, top_k, temperature, max_tokens):