Update app.py

app.py

@@ -6,25 +6,41 @@ app = FastAPI()
 
 pipe = pipeline(
     "text-generation",
-    model="TinyLlama/TinyLlama-1.1B-Chat-v1.0"
+    model="microsoft/phi-3-mini-4k-instruct",
+    device_map="auto"
 )
 
 class Req(BaseModel):
     message: str
 
+def build_prompt(user_msg: str) -> str:
+    return (
+        "<|system|>\n"
+        "You are a helpful, clear, and concise assistant. "
+        "Give accurate and well-structured answers. Avoid unnecessary fluff."
+        "<|end|>\n"
+        "<|user|>\n"
+        f"{user_msg}"
+        "<|end|>\n"
+        "<|assistant|>\n"
+    )
+
 @app.post("/chat")
 def chat(req: Req):
-    prompt = f"<s>[INST] {req.message} [/INST]"
+    prompt = build_prompt(req.message)
 
     out = pipe(
         prompt,
-        max_new_tokens=100,
-        temperature=0.7
+        max_new_tokens=200,
+        temperature=0.6,
+        top_p=0.9,
+        do_sample=True,
+        repetition_penalty=1.1,
+        eos_token_id=pipe.tokenizer.eos_token_id
     )
 
     text = out[0]["generated_text"]
 
-    # remove prompt from output
-    reply = text.split("[/INST]")[-1].strip()
+    reply = text.split("<|assistant|>")[-1].strip()
 
     return {"reply": reply}