Update app.py

app.py

@@ -77,7 +77,7 @@ def run_training(hf_token, model_name, new_repo_name, lora_r, lora_alpha, lora_d
     def load_all_datasets():
         streams = []
         tasks = []
-        progress(0.1, desc="Analizando configuraciones de datasets...")
+        progress(0.1, desc="Analizando configuraciones...")
         
         for ds_name in dataset_list:
             try:
@@ -91,7 +91,7 @@ def run_training(hf_token, model_name, new_repo_name, lora_r, lora_alpha, lora_d
                 for c in configs:
                     tasks.append((ds_name, c))
 
-        progress(0.2, desc=f"Cargando {len(tasks)} fuentes de datos...")
+        progress(0.2, desc=f"Cargando {len(tasks)} fuentes...")
         with ThreadPoolExecutor(max_workers=num_workers) as executor:
             future_to_task = {executor.submit(load_single, d, c): (d, c) for d, c in tasks}
             for future in as_completed(future_to_task):
@@ -105,14 +105,17 @@ def run_training(hf_token, model_name, new_repo_name, lora_r, lora_alpha, lora_d
 
     loaded_streams = load_all_datasets()
     if not loaded_streams:
-        return "Error Crítico: No se pudo cargar ningún dataset válido."
+        return "Error: No se pudo cargar ningún dataset válido."
 
     def all_samples():
         return chain.from_iterable(loaded_streams)
 
     progress(0.3, desc="Cargando Tokenizer...")
-    tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True, padding_side="left", add_eos_token=True, add_bos_token=True)
-    tokenizer.pad_token = tokenizer.eos_token
+    try:
+        tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True, padding_side="left", add_eos_token=True, add_bos_token=True)
+        tokenizer.pad_token = tokenizer.eos_token
+    except Exception as e:
+        return f"Error cargando tokenizer: {str(e)}"
 
     def create_text_lines(sample):
         if isinstance(sample, dict):
@@ -157,8 +160,11 @@ def run_training(hf_token, model_name, new_repo_name, lora_r, lora_alpha, lora_d
                     except:
                         pass
 
-    progress(0.4, desc="Cargando Modelo Base...")
-    original_model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True).to(device)
+    progress(0.4, desc="Cargando Modelo...")
+    try:
+        original_model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True).to(device)
+    except Exception as e:
+        return f"Error cargando modelo: {str(e)}"
     
     peft_config = LoraConfig(
         r=int(lora_r),
@@ -202,22 +208,22 @@ def run_training(hf_token, model_name, new_repo_name, lora_r, lora_alpha, lora_d
     progress(0.5, desc="Entrenando...")
     trainer.train()
     
-    progress(0.8, desc="Guardando checkpoint...")
+    progress(0.8, desc="Guardando...")
     trainer.save_model(output_dir)
 
-    progress(0.9, desc="Fusionando modelo LoRA...")
+    progress(0.9, desc="Fusionando...")
     ft = PeftModel.from_pretrained(original_model, output_dir, torch_dtype=torch.float32, is_trainable=False, device_map={"": device}).merge_and_unload()
     
     final_path = "/content/merged_model"
     ft.save_pretrained(final_path, safe_serialization=True)
     tokenizer.save_pretrained(final_path)
 
-    progress(0.95, desc="Subiendo a HuggingFace...")
+    progress(0.95, desc="Subiendo...")
     full_repo = f"{username}/{new_repo_name}"
     create_repo(full_repo, token=hf_token, exist_ok=True)
     upload_folder(folder_path=final_path, repo_id=full_repo, token=hf_token)
 
-    return f"¡Éxito! Modelo disponible en: https://huggingface.co/{full_repo}"
+    return f"Completado: https://huggingface.co/{full_repo}"
 
 custom_css = """
 body {background-color: #0b0f19; color: #e0e6ed;}
@@ -240,8 +246,8 @@ with gr.Blocks(css=custom_css, title="Entrenador LLM Ultimate") as demo:
     
     with gr.Row():
         with gr.Column(scale=1):
-            hf_token_input = gr.Textbox(label="HuggingFace Token (Write)", type="password", placeholder="hf_...", value=os.getenv("HF_TOKEN", ""))
-            model_input = gr.Textbox(label="Modelo Base", value="arnir0/Tiny-LLM")
+            hf_token_input = gr.Textbox(label="HuggingFace Token", type="password", placeholder="hf_...", value=os.getenv("HF_TOKEN", ""))
+            model_input = gr.Textbox(label="Modelo Base", value="", placeholder="Ej: Qwen/Qwen2.5-0.5B")
             repo_input = gr.Textbox(label="Nombre Nuevo Repo", value="multi-dataset-model-v1")
         
         with gr.Column(scale=1):
@@ -258,7 +264,7 @@ with gr.Blocks(css=custom_css, title="Entrenador LLM Ultimate") as demo:
 
     datasets_input = gr.Textbox(label="Fuentes de Datos (Datasets)", value="", placeholder="Pega aquí tus datasets separados por coma o salto de línea.\nEjemplo:\nSalesforce/fineweb_deduplicated\nbigcode/the-stack, v2", lines=12, elem_classes="input-box")
     
-    train_btn = gr.Button("🚀 INICIAR ENTRENAMIENTO GLOBAL", elem_classes="primary-btn")
+    train_btn = gr.Button("🚀 INICIAR ENTRENAMIENTO", elem_classes="primary-btn")
     status_output = gr.Textbox(label="Log del Sistema", interactive=False, lines=3)
 
     train_btn.click(