Create app.py

app.py

@@ -0,0 +1,65 @@
+from fastapi import FastAPI
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
+import os
+from pydantic import BaseModel
+
+# ✅ Force Hugging Face cache to /tmp (writable in Spaces)
+os.environ["HF_HOME"] = "/tmp"
+os.environ["TRANSFORMERS_CACHE"] = "/tmp"
+
+
+model_id = "rabiyulfahim/qa_python_gpt2"
+
+tokenizer = AutoTokenizer.from_pretrained(model_id, cache_dir="/tmp")
+model = AutoModelForCausalLM.from_pretrained(model_id, cache_dir="/tmp")
+
+
+app = FastAPI(title="QA GPT2 API", description="Serving HuggingFace model with FastAPI")
+
+
+# Request schema
+class QueryRequest(BaseModel):
+    question: str
+    max_new_tokens: int = 50
+    temperature: float = 0.7
+    top_p: float = 0.9
+
+
+@app.get("/")
+def home():
+    return {"message": "Welcome to QA GPT2 API 🚀"}
+
+@app.get("/ask")
+def ask(question: str, max_new_tokens: int = 50):
+    inputs = tokenizer(question, return_tensors="pt")
+    outputs = model.generate(**inputs, max_new_tokens=max_new_tokens)
+    answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return {"question": question, "answer": answer}
+
+
+
+# Health check endpoint
+@app.get("/health")
+def health():
+    return {"status": "ok"}
+
+# Inference endpoint
+@app.post("/predict")
+def predict(request: QueryRequest):
+    inputs = tokenizer(request.question, return_tensors="pt")
+    outputs = model.generate(
+        **inputs,
+        max_new_tokens=request.max_new_tokens,
+        do_sample=True,
+        temperature=0.7,
+        top_p=0.9,
+        pad_token_id=tokenizer.eos_token_id,
+        return_dict_in_generate=True
+    )
+
+    answer = tokenizer.decode(outputs.sequences[0], skip_special_tokens=True)
+    return {
+        "question": request.question,
+        "answer": answer
+    }