Update inference.py

inference.py

@@ -61,4 +61,15 @@ def hf_inference_fn(inputs: str):
         embedding_src=embedding_matrix_en,
         device=device,
         max_len=max_len
-    )
+    )
+from fastapi import FastAPI
+from pydantic import BaseModel
+
+app = FastAPI()
+
+class TranslateRequest(BaseModel):
+    text: str
+
+@app.post("/translate")
+def translate(req: TranslateRequest):
+    return {"translation": hf_inference_fn(req.text)}