Create app.py

app.py

@@ -0,0 +1,171 @@
+import gradio as gr
+import torch
+from transformers import pipeline
+from thinking_btn import CUSTOM_CSS, THINKING_JS
+from logo import get_logo_html
+
+pipe = pipeline(
+    "text-generation",
+    model="Qwen/Qwen2-1.5B-Instruct",
+    model_kwargs={"torch_dtype": torch.float16 if torch.cuda.is_available() else torch.float32},
+    device_map="auto",
+)
+
+def chat(message, history):
+    if not message or not message.strip():
+        return ""
+    try:
+        if len(message) > 1000:
+            message = message[:1000] + "... [texto cortado]"
+
+        system_msg = (
+            "Você é BrainGPT — uma inteligência artificial poderosa, direta e precisa. "
+            "Responda sempre em português de forma clara, concisa e inteligente. "
+            "Nunca invente factos. Se não souber, diga claramente."
+        )
+
+        msgs = [{"role": "system", "content": system_msg}]
+        for user_msg, bot_msg in history[-6:]:
+            if user_msg:
+                msgs.append({"role": "user", "content": user_msg})
+            if bot_msg:
+                msgs.append({"role": "assistant", "content": bot_msg})
+        msgs.append({"role": "user", "content": message})
+
+        prompt = pipe.tokenizer.apply_chat_template(
+            msgs, tokenize=False, add_generation_prompt=True
+        )
+
+        if len(prompt) > 3500:
+            return "Conversa muito longa. Clica em Limpar e recomeça."
+
+        result = pipe(
+            prompt,
+            max_new_tokens=280,
+            do_sample=True,
+            temperature=0.7,
+            top_p=0.9,
+            pad_token_id=pipe.tokenizer.eos_token_id,
+            truncation=True,
+        )
+
+        full = result[0]["generated_text"]
+        resposta = full.split("<|im_start|>assistant\n")[-1].strip()
+        resposta = resposta.replace("<|im_end|>", "").strip()
+
+        return resposta if resposta else "Resposta vazia. Tenta reformular."
+
+    except torch.cuda.OutOfMemoryError:
+        return "Memoria GPU esgotada. Clica em Limpar e tenta novamente."
+    except Exception as e:
+        return f"Erro: {str(e)[:150]}"
+
+
+with gr.Blocks(
+    css=CUSTOM_CSS,
+    js=THINKING_JS,
+    title="BrainGPT",
+    theme=gr.themes.Base(
+        primary_hue="orange",
+        neutral_hue="stone",
+    ),
+) as demo:
+
+    gr.HTML(get_logo_html("logo.png"))
+
+    gr.HTML("""
+    <div id="thinking-bar" class="thinking-bar">
+        <div class="thinking-dots">
+            <span></span><span></span><span></span>
+        </div>
+        <span>THINKING...</span>
+    </div>
+    """)
+
+    chatbot = gr.Chatbot(
+        label="",
+        height=480,
+        show_copy_button=True,
+        bubble_full_width=False,
+        avatar_images=(None, "logo.png"),
+    )
+
+    with gr.Row(elem_classes="input-row"):
+        msg = gr.Textbox(
+            placeholder="Escreve a tua mensagem...",
+            show_label=False,
+            lines=1,
+            max_lines=5,
+            elem_id="msg-input",
+            scale=8,
+        )
+        send_btn = gr.Button(
+            "Enviar",
+            elem_id="send-btn",
+            variant="primary",
+            scale=2,
+        )
+        clear_btn = gr.Button(
+            "Limpar",
+            elem_id="clear-btn",
+            variant="secondary",
+            scale=1,
+        )
+
+    gr.Examples(
+        examples=[
+            "Quem es tu?",
+            "Explica inteligencia artificial em 3 frases",
+            "Diferenca entre machine learning e deep learning?",
+            "5 dicas para ser mais produtivo",
+            "Como funciona o Bitcoin?",
+        ],
+        inputs=msg,
+        label="Exemplos",
+    )
+
+    def user_submit(message, history):
+        if not message.strip():
+            return "", history
+        history = history + [[message, None]]
+        return "", history
+
+    def bot_respond(history):
+        if not history or history[-1][1] is not None:
+            return history
+        message = history[-1][0]
+        response = chat(message, history[:-1])
+        history[-1][1] = response
+        return history
+
+    msg.submit(
+        user_submit,
+        inputs=[msg, chatbot],
+        outputs=[msg, chatbot],
+        queue=False,
+    ).then(
+        bot_respond,
+        inputs=[chatbot],
+        outputs=[chatbot],
+    )
+
+    send_btn.click(
+        user_submit,
+        inputs=[msg, chatbot],
+        outputs=[msg, chatbot],
+        queue=False,
+    ).then(
+        bot_respond,
+        inputs=[chatbot],
+        outputs=[chatbot],
+    )
+
+    clear_btn.click(
+        lambda: ([], ""),
+        outputs=[chatbot, msg],
+        queue=False,
+    )
+
+if __name__ == "__main__":
+    demo.queue(max_size=5).launch()
+