Update app.py

app.py

@@ -3,20 +3,14 @@ from PIL import Image
 from transformers import AutoModel, AutoTokenizer
 import gradio as gr
 
-# Configuração inicial
-torch.manual_seed(100)
-
 # Carregar o modelo e o tokenizer
-model = AutoModel.from_pretrained('openbmb/MiniCPM-V', trust_remote_code=True,
-                                  attn_implementation='sdpa', torch_dtype=torch.bfloat16)
-model = model.eval().cuda()
+model = AutoModel.from_pretrained('openbmb/MiniCPM-V', trust_remote_code=True, torch_dtype=torch.bfloat16)
+model = model.to(device='cuda', dtype=torch.bfloat16)  # Ajuste para o dispositivo e tipo de dados adequados
 tokenizer = AutoTokenizer.from_pretrained('openbmb/MiniCPM-V', trust_remote_code=True)
+model.eval()
 
-# Função para interagir com o modelo
-def chat_with_model(image, question, chat_history=None):
-    if chat_history is None:
-        chat_history = []
-
+# Função para processar a imagem e a pergunta
+def chat_with_model(image, question):
     # Converter a imagem para RGB (se necessário)
     if isinstance(image, str):
         image = Image.open(image).convert('RGB')
@@ -24,49 +18,43 @@ def chat_with_model(image, question, chat_history=None):
         image = image.convert('RGB')
 
     # Preparar a mensagem para o modelo
-    msgs = [{'role': 'user', 'content': [image, question]}]
-
-    # Adicionar histórico de conversa, se houver
-    for msg in chat_history:
-        msgs.append(msg)
+    msgs = [{'role': 'user', 'content': question}]
 
     # Gerar resposta do modelo
-    answer = model.chat(
+    res, context, _ = model.chat(
+        image=image,
         msgs=msgs,
-        tokenizer=tokenizer
+        context=None,
+        tokenizer=tokenizer,
+        sampling=True,
+        temperature=0.7
     )
 
-    # Atualizar o histórico de conversa
-    chat_history.append({"role": "user", "content": [image, question]})
-    chat_history.append({"role": "assistant", "content": [answer]})
-
-    # Retornar a resposta e o histórico atualizado
-    return answer, chat_history
+    return res
 
 # Interface Gradio
-def gradio_interface(image, question, chat_history=None):
-    response, updated_history = chat_with_model(image, question, chat_history)
-    return response, updated_history
+def gradio_interface(image, question):
+    response = chat_with_model(image, question)
+    return response
 
 # Criar a interface Gradio
 with gr.Blocks() as demo:
-    gr.Markdown("# MiniCPM-o-2_6 Chat with Images")
+    gr.Markdown("# MiniCPM-V Chat with Images")
     gr.Markdown("Envie uma imagem e faça perguntas sobre ela.")
 
     with gr.Row():
-        image_input = gr.Image(label="Upload Image", type="pil")
-        question_input = gr.Textbox(label="Your Question", placeholder="What is in the image?")
+        image_input = gr.Image(label="Upload Image", type="pil")  # Campo para upload de imagem
+        question_input = gr.Textbox(label="Your Question", placeholder="What is in the image?")  # Campo para a pergunta
     
-    chat_history = gr.State([])  # Armazenar o histórico de conversa
-    output_text = gr.Textbox(label="Model Response", interactive=False)
+    output_text = gr.Textbox(label="Model Response", interactive=False)  # Campo para exibir a resposta
     
-    submit_button = gr.Button("Submit")
+    submit_button = gr.Button("Submit")  # Botão para enviar
     
     # Ação ao clicar no botão
     submit_button.click(
         fn=gradio_interface,
-        inputs=[image_input, question_input, chat_history],
-        outputs=[output_text, chat_history]
+        inputs=[image_input, question_input],
+        outputs=output_text
     )
 
 # Iniciar a interface