Update app.py

app.py

@@ -1,27 +1,32 @@
+import os
+os.environ["HF_HOME"] = "/tmp/hf"  # cache scriibil în Space
+
 from fastapi import FastAPI, Request
 import json
 import faiss
 from sentence_transformers import SentenceTransformer
-from transformers import AutoTokenizer, AutoModelForCausalLM
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 import torch
 
 app = FastAPI()
 
 # ---------------------------
-# 1. Încarcă modelul Gemma 1B
+# 1. Încarcă modelul
 # ---------------------------
-MODEL_NAME = "distilgpt2"  # modelul mic Gemma 1B
+MODEL_NAME = "google/flan-t5-small"  # public și mic
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
+model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME)
 
 # ---------------------------
 # 2. Încarcă articolele și embeddings
 # ---------------------------
-with open("articles.json", "r") as f:
+with open("articles.json", "r", encoding="utf-8") as f:
     articles = json.load(f)
 
+# fiecare articol -> text
 sentences = [a["content"] for a in articles]
 
+# embeddings rapide
 embedder = SentenceTransformer("all-MiniLM-L6-v2")
 embeddings = embedder.encode(sentences)
 index = faiss.IndexFlatL2(embeddings.shape[1])
@@ -34,7 +39,7 @@ index.add(embeddings)
 async def ask(request: Request):
     data = await request.json()
     question = data.get("question", "")
-    
+
     # căutare semantică
     q_emb = embedder.encode([question])
     D, I = index.search(q_emb, k=3)
@@ -46,7 +51,7 @@ async def ask(request: Request):
     inputs = tokenizer(prompt, return_tensors="pt")
     outputs = model.generate(**inputs, max_new_tokens=150)
     answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    
+
     return {"answer": answer}
 
 # ---------------------------