Update app.py

app.py

@@ -1,40 +1,57 @@
-import gradio as gr
-import os
-import requests
+from fastapi import FastAPI, Request
+import json
+import faiss
+from sentence_transformers import SentenceTransformer
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
 
-# Model public gratuit pentru HF Inference API
-MODEL_ID = "distilbert-base-uncased"
+app = FastAPI()
 
-HF_API_TOKEN = os.getenv("HF_API_TOKEN")  # poate fi gol pentru public
-HF_API_URL = f"https://api-inference.huggingface.co/models/{MODEL_ID}"
-HEADERS = {"Authorization": f"Bearer {HF_API_TOKEN}"} if HF_API_TOKEN else {}
+# ---------------------------
+# 1. Încarcă modelul Gemma 1B
+# ---------------------------
+MODEL_NAME = "google/gemma-3-1b-it"  # modelul mic Gemma 1B
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
 
-def ask_ai(question):
-    if not question.strip():
-        return "Te rog scrie o întrebare."
-    
-    payload = {
-        "inputs": question,
-        "parameters": {"max_new_tokens": 150, "return_full_text": False}
-    }
+# ---------------------------
+# 2. Încarcă articolele și embeddings
+# ---------------------------
+with open("articles.json", "r") as f:
+    articles = json.load(f)
+
+sentences = [a["content"] for a in articles]
+
+embedder = SentenceTransformer("all-MiniLM-L6-v2")
+embeddings = embedder.encode(sentences)
+index = faiss.IndexFlatL2(embeddings.shape[1])
+index.add(embeddings)
+
+# ---------------------------
+# 3. Endpoint pentru întrebări
+# ---------------------------
+@app.post("/ask")
+async def ask(request: Request):
+    data = await request.json()
+    question = data.get("question", "")
     
-    try:
-        r = requests.post(HF_API_URL, headers=HEADERS, json=payload, timeout=30)
-        r.raise_for_status()
-    except Exception as e:
-        return f"Eroare la apel HF Inference API: {str(e)}"
+    # căutare semantică
+    q_emb = embedder.encode([question])
+    D, I = index.search(q_emb, k=3)
+    context = " ".join([sentences[i] for i in I[0]])
+
+    # prompt pentru model
+    prompt = f"Context: {context}\nÎntrebare: {question}\nRăspuns:"
+
+    inputs = tokenizer(prompt, return_tensors="pt")
+    outputs = model.generate(**inputs, max_new_tokens=150)
+    answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
     
-    res = r.json()
-    if isinstance(res, list) and "generated_text" in res[0]:
-        return res[0]["generated_text"]
-    return str(res)
-
-iface = gr.Interface(
-    fn=ask_ai,
-    inputs=gr.Textbox(lines=2, placeholder="Întreabă ceva..."),
-    outputs="text",
-    title="Chatbot HF API",
-    description="Chatbot funcțional fără descărcarea modelelor, rulând HF Inference API."
-)
-
-iface.launch(server_name="0.0.0.0", server_port=7860)
+    return {"answer": answer}
+
+# ---------------------------
+# 4. Run server
+# ---------------------------
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run(app, host="0.0.0.0", port=7860)