Update app.py

app.py

@@ -4,13 +4,18 @@ import gradio as gr
 from transformers import AutoTokenizer, AutoModelForCausalLM
 from peft import PeftModel
 
+# =========================
+# CONFIG
+# =========================
 BASE_MODEL = "google/gemma-3-270m-it"
 LORA_MODEL = "loboGOAT/DogeAI-v1.0-instruct"
 
 HF_TOKEN = os.environ.get("HF_TOKEN")
-
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
+# =========================
+# LOAD MODEL
+# =========================
 print("🔄 Loading tokenizer...")
 tokenizer = AutoTokenizer.from_pretrained(
     BASE_MODEL,
@@ -29,10 +34,14 @@ print("🔄 Applying LoRA...")
 model = PeftModel.from_pretrained(model, LORA_MODEL)
 model.eval()
 
-
-def chat(user_input):
+# =========================
+# CHAT FUNCTION
+# =========================
+def chat(user_input, system_prompt, temperature, top_p, max_tokens):
     prompt = (
         "<bos>\n"
+        "<start_of_turn>system\n"
+        f"{system_prompt}\n"
         "<start_of_turn>user\n"
         f"{user_input}\n"
         "<start_of_turn>model\n"
@@ -43,10 +52,10 @@ def chat(user_input):
     with torch.no_grad():
         output = model.generate(
             **inputs,
-            max_new_tokens=200,
+            max_new_tokens=int(max_tokens),
             do_sample=True,
-            temperature=0.7,
-            top_p=0.9,
+            temperature=float(temperature),
+            top_p=float(top_p),
             repetition_penalty=1.1,
             eos_token_id=tokenizer.eos_token_id,
             pad_token_id=tokenizer.pad_token_id
@@ -59,13 +68,73 @@ def chat(user_input):
 
     return text.strip()
 
-demo = gr.Interface(
-    fn=chat,
-    inputs=gr.Textbox(lines=4, placeholder="Talk to DogeAI 🐕"),
-    outputs="text",
-    title="DogeAI v1.0",
-    description="LoRA fine-tuned Gemma-3-270M-it • Created by Arthur"
-)
+# =========================
+# GRADIO UI
+# =========================
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    gr.Markdown(
+        """
+        # 🐕 DogeAI v1.0  
+
+        ⚠️ **AVISO IMPORTANTE**  
+        Este modelo é **experimental**, pequeno e pode **ALUCINAR** respostas,  
+        **inventar fatos**, errar datas, nomes ou explicações técnicas.
+
+        ❌ **NÃO use como fonte confiável**  
+        ✅ Use para estudo, testes e experimentação
+        """
+    )
+
+    with gr.Row():
+        with gr.Column(scale=3):
+            user_input = gr.Textbox(
+                lines=5,
+                label="Mensagem",
+                placeholder="Fale com o DogeAI 🐶"
+            )
+
+            submit = gr.Button("Enviar 🚀")
+
+            output = gr.Textbox(
+                lines=10,
+                label="Resposta do modelo"
+            )
+
+        with gr.Column(scale=2):
+            gr.Markdown("### 🧠 System Prompt")
+            system_prompt = gr.Textbox(
+                lines=6,
+                value=(
+                    "Você é o DogeAI, um modelo experimental. "
+                    "Seja honesto quando não souber algo. "
+                    "Não invente fatos."
+                ),
+                label="Instruções internas"
+            )
+
+            gr.Markdown("### ⚠️ Hiperparâmetros")
+            gr.Markdown(
+                "**SÓ MEXA SE SOUBER O QUE ESTÁ FAZENDO**/**ONLY ADJUST VALUES IF YOU KNOW WHAT YOU'RE DOING**\n\n"
+                "Valores errados podem causar respostas ruins ou nonsense./Bad values can cause trash or nonsense responses"
+            )
+
+            temperature = gr.Slider(
+                0.1, 1.5, value=0.7, step=0.05, label="Temperature"
+            )
+
+            top_p = gr.Slider(
+                0.1, 1.0, value=0.9, step=0.05, label="Top-p"
+            )
+
+            max_tokens = gr.Slider(
+                50, 512, value=200, step=10, label="Max tokens"
+            )
+
+    submit.click(
+        chat,
+        inputs=[user_input, system_prompt, temperature, top_p, max_tokens],
+        outputs=output
+    )
 
 demo.launch()