Create app.py

app.py

@@ -0,0 +1,35 @@
+from fastapi import FastAPI, Request
+from transformers import pipeline
+
+app = FastAPI()
+
+generator = pipeline("text-generation", model="microsoft/phi-2", max_new_tokens=150)
+
+sessions = {}
+
+@app.get("/")
+def read_root():
+    return {"message": "LLM API running!"}
+
+@app.post("/chat")
+async def chat(request: Request):
+    data = await request.json()
+    session_id = data.get("session_id", "default")
+    user_input = data.get("message", "")
+    system_prompt = data.get("system_prompt", "You are a helpful assistant.")
+
+    if session_id not in sessions:
+        sessions[session_id] = system_prompt + "\n"
+
+    sessions[session_id] += f"User: {user_input}\nAssistant:"
+
+    output = generator(sessions[session_id], max_new_tokens=150)[0]["generated_text"]
+
+    if "Assistant:" in output:
+        assistant_response = output.split("Assistant:")[-1].strip()
+    else:
+        assistant_response = output.strip()
+
+    sessions[session_id] += f" {assistant_response}\n"
+
+    return {"response": assistant_response}