Create app.py

app.py

@@ -0,0 +1,19 @@
+from fastapi import FastAPI, Request
+from transformers import pipeline
+
+app = FastAPI()
+
+pipe = pipeline(
+    "text-generation",
+    model="google/gemma-3-1b-it",
+    device="cpu",
+    max_new_tokens=256,
+    temperature=0.7
+)
+
+@app.post("/generate")
+async def generate(request: Request):
+    data = await request.json()
+    prompt = data.get("prompt", "")
+    result = pipe(prompt)
+    return {"response": result[0]["generated_text"]}