Update app.py

app.py

@@ -1,109 +1,34 @@
-# app.py
-import os
-import json
-import faiss
-import numpy as np
-import requests
-from fastapi import FastAPI, HTTPException
-from pydantic import BaseModel
-from sentence_transformers import SentenceTransformer
-
-# config
-JSON_FILE = "articles.json"
-EMBED_MODEL = "sentence-transformers/all-MiniLM-L6-v2"
-TOP_K = 3
-HF_MODEL = os.getenv("HF_MODEL", "google/flan-t5-small")    # model hosted on HF
-HF_TOKEN = os.getenv("HF_API_TOKEN")  # set in Secrets on HF Spaces
-SERVER_PORT = int(os.getenv("PORT", 7860))
-
-# load embeddings
-embed_model = SentenceTransformer(EMBED_MODEL)
-
-# build index
-if not os.path.exists(JSON_FILE):
-    raise FileNotFoundError("articles.json not found")
-with open(JSON_FILE, "r", encoding="utf-8") as f:
-    articles = json.load(f)
-
-def chunk_text(text, size=500, overlap=100):
-    chunks=[]
-    s=0
-    while s < len(text):
-        e=min(s+size, len(text))
-        chunks.append(text[s:e])
-        s=e-overlap
-        if s<0: s=0
-        if s>=len(text): break
-    return chunks
-
-texts=[]; metas=[]; embs_list=[]
-for i,art in enumerate(articles):
-    content = art.get("continut") or art.get("Continut") or ""
-    if not content.strip(): continue
-    url = art.get("url") or art.get("URL") or ""
-    title = art.get("titlu") or art.get("Titlu") or f"art_{i}"
-    chunks = chunk_text(content)
-    if not chunks: continue
-    embs = embed_model.encode(chunks, convert_to_numpy=True)
-    if embs.ndim==1: embs = embs.reshape(1,-1)
-    embs_list.append(embs)
-    texts.extend(chunks)
-    metas.extend([{"title":title,"url":url,"chunk":j} for j in range(len(chunks))])
-
-if len(embs_list)==0:
-    raise ValueError("No valid chunks in articles.json")
-
-embeddings = np.vstack(embs_list).astype("float32")
-faiss.normalize_L2(embeddings)
-index = faiss.IndexFlatIP(embeddings.shape[1])
-index.add(embeddings)
-metadata={"texts":texts,"metas":metas}
-
-# HF generation helper
-def generate_via_hf(prompt, max_tokens=128):
-    if not HF_TOKEN:
-        raise RuntimeError("HF_API_TOKEN not set in env")
-    url = f"https://api-inference.huggingface.co/models/{HF_MODEL}"
-    headers = {"Authorization": f"Bearer {HF_TOKEN}"}
-    payload = {"inputs": prompt, "parameters": {"max_new_tokens": max_tokens, "do_sample": False}}
-    r = requests.post(url, headers=headers, json=payload, timeout=60)
-    r.raise_for_status()
-    data = r.json()
-    # handle expected response
-    if isinstance(data, list) and "generated_text" in data[0]:
-        return data[0]["generated_text"]
-    if isinstance(data, dict) and "error" in data:
-        raise RuntimeError("HF error: " + data["error"])
-    return str(data)
-
-# FastAPI
-app = FastAPI()
-
-class Q(BaseModel):
-    question: str
-
-@app.get("/ping")
-def ping():
-    return {"status":"ok"}
-
-@app.post("/ask")
-def ask(q: Q):
-    qtext = q.question.strip()
-    if not qtext:
-        raise HTTPException(status_code=400, detail="Empty question")
-    q_emb = embed_model.encode([qtext], convert_to_numpy=True).astype("float32")
-    if q_emb.ndim==1: q_emb = q_emb.reshape(1,-1)
-    faiss.normalize_L2(q_emb)
-    k = min(TOP_K, index.ntotal)
-    if k<=0:
-        return {"answer":"No articles indexed."}
-    D,I = index.search(q_emb, k)
-    retrieved = [metadata["texts"][i] for i in I[0]]
-    urls = [metadata["metas"][i].get("url","") for i in I[0]]
-    context = "\n\n".join(retrieved)
-    prompt = f"Context:\n{context}\n\nQuestion: {qtext}\nAnswer:"
-    try:
-        generated = generate_via_hf(prompt, max_tokens=128)
-    except Exception as e:
-        return {"answer": f"HF generation error: {e}", "sources": urls}
-    return {"answer": generated, "sources": [u for u in urls if u]}
+import gradio as gr
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
+import torch
+
+# Model public, mic și gratuit
+MODEL_NAME = "google/flan-t5-small"
+
+# Încarcă model și tokenizer
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME)
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+model = model.to(device)
+
+# Funcția chatbot
+def chat_fn(question):
+    if not question.strip():
+        return "Te rog scrie o întrebare."
+    
+    inputs = tokenizer(question, return_tensors="pt").to(device)
+    outputs = model.generate(**inputs, max_new_tokens=150)
+    answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return answer
+
+# Gradio UI
+iface = gr.Interface(
+    fn=chat_fn,
+    inputs=gr.Textbox(lines=2, placeholder="Întreabă ceva..."),
+    outputs="text",
+    title="Chatbot simplu",
+    description="Chatbot minimal bazat pe Flan-T5-small (fără date pre-trained locale)."
+)
+
+iface.launch(server_name="0.0.0.0", server_port=7860)