Update app.py

app.py

@@ -3,31 +3,38 @@ import pandas as pd
 import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
 
-tokenizer = AutoTokenizer.from_pretrained("distilgpt2")
-model = AutoModelForCausalLM.from_pretrained("distilgpt2", torch_dtype=torch.float16)
-model = model.to('cuda') if torch.cuda.is_available() else model.to('cpu')
+# Load the tokenizer and quantized model
+model_name = "meta-llama/Meta-Llama-3.1-8B"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+# Use bitsandbytes to load the model in 8-bit precision
+model = AutoModelForCausalLM.from_pretrained(model_name, load_in_8bit=True, device_map='auto')
+
+# Move model to the appropriate device (GPU/CPU)
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+model = model.to(device)
 
 # Set the padding token to the end-of-sequence token
 if tokenizer.pad_token is None:
     tokenizer.pad_token = tokenizer.eos_token
 
-df = pd.read_csv('anomalies.csv')
+# Load the anomalies data
+df = pd.read_csv('anomalies.csv', sep=',', decimal='.')
 
-# Função para gerar resposta
+# Function to generate a response
 def response(question):
-    prompt = f"Considerando os dados: {df.to_string(index=False)}, onde 'ds' está em formato DateTime, 'real' é o valor da despesa e 'group' é o grupo da despesa. Pergunta: {question}"
-    inputs = tokenizer(prompt, return_tensors='pt', padding='max_length', truncation=True, max_length=256)
-    attention_mask = inputs['attention_mask']
-    input_ids = inputs['input_ids']
+    prompt = f"Considerando os dados: {df.to_string(index=False)}, onde a coluna 'ds' está em formato DateTime, a coluna 'real' é o valor da despesa e a coluna 'group' é o grupo da despesa. Pergunta: {question}"
+    
+    inputs = tokenizer(prompt, return_tensors='pt', padding='max_length', truncation=True, max_length=256).to(device)
     
     generated_ids = model.generate(
-        input_ids,
-        attention_mask=attention_mask,
-        max_length=len(input_ids[0]) + 50,  # Reduce max_length to speed up response
+        inputs['input_ids'],
+        attention_mask=inputs['attention_mask'],
+        max_length=inputs['input_ids'].shape[1] + 50,
         temperature=0.7,
         top_p=0.9,
         no_repeat_ngram_size=2,
-        num_beams=3,  # Adding beams for more reliable generation
+        num_beams=3,
     )
     
     generated_text = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
@@ -35,7 +42,7 @@ def response(question):
     
     return final_response
 
-# Interface Streamlit
+# Streamlit interface
 st.markdown("""
 <div style='display: flex; align-items: center;'>
     <div style='width: 40px; height: 40px; background-color: green; border-radius: 50%; margin-right: 5px;'></div>
@@ -45,30 +52,30 @@ st.markdown("""
 </div>
 """, unsafe_allow_html=True)
 
-# Histórico de conversas
+# Chat history
 if 'history' not in st.session_state:
     st.session_state['history'] = []
 
-# Caixa de entrada para a pergunta
+# Input box for user question
 user_question = st.text_input("Escreva sua questão aqui:", "")
 
 if user_question:
-    # Adiciona emoji de pessoa quando a pergunta está sendo digitada
+    # Add person emoji when typing question
     st.session_state['history'].append(('👤', user_question))
     st.markdown(f"**👤 {user_question}**")
     
-    # Gera a resposta
+    # Generate the response
     bot_response = response(user_question)
     
-    # Adiciona emoji de robô quando a resposta está sendo gerada e alinha à direita
+    # Add robot emoji when generating response and align to the right
     st.session_state['history'].append(('🤖', bot_response))
     st.markdown(f"<div style='text-align: right'>**🤖 {bot_response}**</div>", unsafe_allow_html=True)
 
-# Botão para limpar o histórico
+# Clear history button
 if st.button("Limpar"):
     st.session_state['history'] = []
 
-# Exibe o histórico de conversas
+# Display chat history
 for sender, message in st.session_state['history']:
     if sender == '👤':
         st.markdown(f"**👤 {message}**")