Update app.py

app.py

@@ -2,12 +2,23 @@ import os
 from pathlib import Path
 import gradio as gr
 import numpy as np
-import torch
+from sentence_transformers import SentenceTransformer
 import faiss
+from openai import OpenAI, OpenAIError
+from langchain_community.document_loaders import PyPDFLoader
+from langchain.text_splitter import RecursiveCharacterTextSplitter
 
-from sentence_transformers import SentenceTransformer
-from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
-from huggingface_hub import login
+# ----------------------------
+# API Key NVIDIA (defina diretamente aqui)
+# ----------------------------
+NV_API_KEY = "nvapi"
+
+# Cliente NVIDIA para chat
+client = OpenAI(
+    base_url="https://integrate.api.nvidia.com/v1",
+    api_key=NV_API_KEY
+)
+CHAT_MODEL = "meta/llama3-8b-instruct"
 
 # ----------------------------
 # Configurações da aplicação
@@ -31,50 +42,30 @@ SUGGESTION_QUESTIONS = [
     "Certificações?",
 ]
 
-# Caminhos dos arquivos de índice
+# Paths for files generated by build_index.py
 INDEX_FILE  = "r_docs.index"
 CHUNKS_FILE = "r_chunks.npy"
 PDF_PATH    = "CV-Ronaldo_Menezes_2025_06.pdf"
 
-# Verificação dos arquivos
+# Verificação de índices gerados
 if not Path(INDEX_FILE).exists() or not Path(CHUNKS_FILE).exists():
-    raise FileNotFoundError("Index not found. Por favor, execute: python build_index.py")
+    raise FileNotFoundError(
+        "Index not found. Please run first:\n  python build_index.py"
+    )
 
-# Carrega o índice FAISS e os trechos
+# Carrega FAISS index e chunks
 index  = faiss.read_index(INDEX_FILE)
 chunks = np.load(CHUNKS_FILE, allow_pickle=True)
 
-# Modelo de embeddings
+# Embedding model para seleção de contexto
 embedding_model = SentenceTransformer('all-MiniLM-L6-v2')
 
-# ----------------------------
-# Carregamento do modelo LLM local
-# ----------------------------
-hf_token = os.getenv("HF_TOKEN")
-if hf_token is None:
-    raise ValueError("Token Hugging Face não encontrado. Defina como segredo 'HF_TOKEN' nos Settings do Space.")
-
-# Autenticação
-login(token=hf_token)
-
-MODEL_NAME = "microsoft/phi-2"
-tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, token=hf_token)
-model = AutoModelForCausalLM.from_pretrained(
-    MODEL_NAME,
-    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
-    device_map="auto",
-    token=hf_token
-)
-llm_pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer)
-
-# ----------------------------
-# Recuperação de contexto com FAISS
-# ----------------------------
 def retrieve_context(query: str, k: int = 4) -> str:
-    q_emb = embedding_model.encode([query], convert_to_numpy=True, show_progress_bar=False)
+    q_emb = embedding_model.encode([query], convert_to_numpy=True)
     _, I = index.search(q_emb, k)
     return "\n---\n".join(chunks[i] for i in I[0])
 
+# Histórico de diálogo: tuplas (user, assistant)
 dialog_history: list[tuple[str, str]] = []
 
 # ----------------------------
@@ -85,35 +76,53 @@ def chatbot(user_input, temperature, top_p, max_tokens):
     if not user_input:
         return dialog_history, ""
 
+    # Recupera contexto e monta mensagem de sistema
     context = retrieve_context(user_input)
-
-    prompt = (
-        f"### System:\nVocê é um assistente especializado em pacotes R. "
-        f"Use somente o contexto abaixo para responder. Se não souber, diga isso.\n\n"
-        f"=== Retrieved Context ===\n{context}\n\n"
-        f"### User:\n{user_input}\n\n### Assistant:\n"
-    )
-
+    system_msg = {
+        "role": "system",
+        "content": (
+            "You are an assistant specialized in R packages. "
+            "Use only the context below to answer. If you don't know, say so.\n\n"
+            f"=== Retrieved Context ===\n{context}\n\n"
+        )
+    }
+
+    # Constrói lista de mensagens
+    messages = [system_msg]
+    for u, a in dialog_history:
+        messages.append({"role": "user", "content": u})
+        messages.append({"role": "assistant", "content": a})
+    messages.append({"role": "user", "content": user_input})
+
+    # Chama a API NVIDIA em streaming
+    assistant_reply = ""
     try:
-        result = llm_pipeline(
-            prompt[:model.config.max_position_embeddings],
-            max_new_tokens=max_tokens,
+        stream = client.chat.completions.create(
+            model=CHAT_MODEL,
+            messages=messages,
             temperature=temperature,
             top_p=top_p,
-            do_sample=True
-        )[0]
-        assistant_reply = result.get('generated_text') or result.get('text') or "⚠️ Resposta não gerada."
-        assistant_reply = assistant_reply.split("### Assistant:")[-1].strip()
-    except Exception as e:
-        assistant_reply = f"⚠️ Erro local: {e}"
-
+            max_tokens=max_tokens,
+            stream=True
+        )
+        for chunk in stream:
+            delta = chunk.choices[0].delta
+            if hasattr(delta, "content") and delta.content:
+                assistant_reply += delta.content
+    except OpenAIError as e:
+        assistant_reply = f"⚠️ API Error: {e.__class__.__name__}: {e}"
+
+    # Atualiza histórico e retorna
     dialog_history.append((user_input, assistant_reply))
     return dialog_history, ""
 
-def clear_all():
+# ----------------------------
+# Limpa histórico
+# ----------------------------
+def clear_history():
     global dialog_history
     dialog_history = []
-    return [], "", 0.6, 0.95, 512
+    return [], ""
 
 # ----------------------------
 # Interface Gradio
@@ -141,25 +150,23 @@ with gr.Blocks(title=APP_TITLE, css=custom_css, theme=gr.themes.Base()) as demo:
     gr.Markdown(INTRO)
 
     with gr.Row():
+        # Coluna principal de chat
         with gr.Column(scale=3):
-            chatbot_ui = gr.Chatbot(type="tuples", elem_id="chat-window", render_markdown=True)
+            chatbot_ui = gr.Chatbot(type="tuples", elem_id="chat-window")
             with gr.Row(elem_id="input-area"):
                 txt = gr.Textbox(placeholder="Digite sua pergunta…", lines=2, elem_id="user-input")
                 btn = gr.Button("Enviar", elem_id="send-button")
-            temp = gr.Slider(0, 1, 0.6, label="Temperatura")
-            topp = gr.Slider(0, 1, 0.95, label="Top-p")
-            maxtok = gr.Slider(64, 1024, 512, label="Tokens Máximos")
-
-            btn.click(chatbot, [txt, temp, topp, maxtok], [chatbot_ui, txt])
-            txt.submit(chatbot, [txt, temp, topp, maxtok], [chatbot_ui, txt])
-            gr.Button("Limpar").click(clear_all, [], [chatbot_ui, txt, temp, topp, maxtok])
+            btn.click(chatbot, [txt, gr.Slider(0, 1, 0.6), gr.Slider(0, 1, 0.95), gr.Slider(64, 2048, 512)], [chatbot_ui, txt])
+            txt.submit(chatbot, [txt, gr.Slider(0, 1, 0.6), gr.Slider(0, 1, 0.95), gr.Slider(64, 2048, 512)], [chatbot_ui, txt])
+            gr.Button("Limpar").click(clear_history, [], [chatbot_ui, txt])
 
+        # Sidebar com PDF e sugestões
         with gr.Column(scale=1, elem_classes="sidebar"):
             if Path(PDF_PATH).exists():
                 gr.Markdown(f"[📄 Baixar CV em PDF](/file={PDF_PATH})")
             gr.Markdown("### Sugestões de Perguntas")
             for q in SUGGESTION_QUESTIONS:
-                gr.Button(q).click(lambda q=q: (q, *chatbot(q, 0.6, 0.95, 512)), outputs=[txt, chatbot_ui, txt])
+                gr.Button(q).click(lambda q=q: q, inputs=[], outputs=[txt])
             gr.Markdown("---")
             gr.Markdown("### Dicas de Exploração do PDF")
             gr.Markdown("• Use palavras-chave como 'Process Mining' ou 'GIS' para ir direto à seção relevante.")
@@ -167,3 +174,4 @@ with gr.Blocks(title=APP_TITLE, css=custom_css, theme=gr.themes.Base()) as demo:
 
 if __name__ == "__main__":
     demo.launch(server_name="0.0.0.0", server_port=7860)
+