app para avalição do modelo treinado com interface

app.py

@@ -1,64 +1,75 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
+import csv
+from datetime import datetime
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+# Load models and tokenizers
+base_model_name = "unsloth/Llama-3.2-1B-Instruct"
+finetuned_model_name = "exo-is/esg-context-llama-1Bst-11M"  # Replace with your model's path
 
+tokenizer = AutoTokenizer.from_pretrained(base_model_name)
+base_model = AutoModelForCausalLM.from_pretrained(base_model_name, torch_dtype=torch.float32, low_cpu_mem_usage=True, device_map="cpu")
+finetuned_model = AutoModelForCausalLM.from_pretrained(finetuned_model_name, torch_dtype=torch.float32, low_cpu_mem_usage=True, device_map="cpu")
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+def generate_text(model, prompt, max_new_tokens, temperature):
+    inputs = tokenizer(prompt, return_tensors="pt")
+    
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=int(max_new_tokens),
+            temperature=temperature,
+            num_return_sequences=1,
+            do_sample=True,
+        )
+    
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+def log_interaction(model, prompt, output, validation):
+    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+    with open('interaction_log.csv', 'a', newline='') as file:
+        writer = csv.writer(file)
+        writer.writerow([timestamp, model, prompt, output, validation])
 
-    messages.append({"role": "user", "content": message})
+def generate_and_compare(prompt, max_new_tokens, temperature):
+    base_output = generate_text(base_model, prompt, max_new_tokens, temperature)
+    finetuned_output = generate_text(finetuned_model, prompt, max_new_tokens, temperature)
+    return base_output, finetuned_output
 
-    response = ""
+def evaluate(model, output, score):
+    log_interaction(model, gr.get_state('last_prompt'), output, score)
+    return f"Avaliação registrada: {score}"
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+with gr.Blocks() as demo:
+    gr.Markdown("# Comparação de Modelos: Llama-3.2-1B-Instruct vs. Modelo Fine-tuned para Sustentabilidade")
+    
+    with gr.Row():
+        prompt = gr.Textbox(lines=5, label="Insira seu prompt aqui")
+        max_new_tokens = gr.Slider(50, 500, value=200, step=1, label="Máximo de Novos Tokens")
+        temperature = gr.Slider(0.1, 2.0, value=0.7, step=0.1, label="Temperatura")
+    
+    generate_btn = gr.Button("Gerar")
+    
+    with gr.Row():
+        with gr.Column():
+            base_output = gr.Textbox(label="Saída do Modelo Base", lines=10)
+            base_rating = gr.Radio(["1", "2", "3", "4", "5"], label="Avalie a resposta do Modelo Base")
+            base_submit = gr.Button("Enviar Avaliação (Base)")
+        
+        with gr.Column():
+            finetuned_output = gr.Textbox(label="Saída do Modelo Fine-tuned", lines=10)
+            finetuned_rating = gr.Radio(["1", "2", "3", "4", "5"], label="Avalie a resposta do Modelo Fine-tuned")
+            finetuned_submit = gr.Button("Enviar Avaliação (Fine-tuned)")
+    
+    base_feedback = gr.Textbox(label="Feedback da Avaliação (Base)")
+    finetuned_feedback = gr.Textbox(label="Feedback da Avaliação (Fine-tuned)")
+    
+    generate_btn.click(generate_and_compare, inputs=[prompt, max_new_tokens, temperature], outputs=[base_output, finetuned_output])
+    base_submit.click(evaluate, inputs=["Base", base_output, base_rating], outputs=base_feedback)
+    finetuned_submit.click(evaluate, inputs=["Fine-tuned", finetuned_output, finetuned_rating], outputs=finetuned_feedback)
+    
+    demo.load(lambda: gr.update(value=""), outputs=[prompt])
+    prompt.change(lambda x: gr.set_state(last_prompt=x), inputs=[prompt])
 
-        response += token
-        yield response
-
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
-
-
-if __name__ == "__main__":
-    demo.launch()
+demo.launch()

requirements.txt

@@ -1 +1,5 @@
-huggingface_hub==0.25.2
+huggingface_hub==0.25.2
+
+gradio
+transformers
+torch