Update app.py

app.py

@@ -1,29 +1,29 @@
+import os
+import torch
 import gradio as gr
 from huggingface_hub import InferenceClient
-import os
-from transformers import pipeline
-import numpy as np
-from model import SAMPLING_RATE, FEATURE_EXTRACTOR
+from model import model_params, AudioDataset
 
 token = os.getenv("HF_TOKEN")
-# modelo = "mixed-data"
-modelo = "cry-detector"
-pipe = pipeline(
-    "audio-classification",
-    model=f"A-POR-LOS-8000/distilhubert-finetuned-{modelo}",
-    use_auth_token=token
-)
+dataset_path = f"data/baby_cry_detection" # PARA MONITOR
+# dataset_path = f"data/mixed_data_nuevo" # PARA CLASIFICADOR
+model, _, _, id2label = model_params(dataset_path)
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")# Usar a GPU o CPU
+model.to(device)# Usar a GPU o CPU
 client = InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct", token=token)
 # client = InferenceClient("mistralai/Mistral-Nemo-Instruct-2407", token=token)
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-    ):
+def predict(audio_path):
+    audio_dataset = AudioDataset(dataset_path, {})
+    inputs = audio_dataset.preprocess_audio(audio_path)
+    inputs = {"input_values": inputs.to(device).unsqueeze(0)}
+    with torch.no_grad():
+        outputs = model(**inputs)
+        predicted_class_ids = outputs.logits.argmax(-1)
+        label = id2label[predicted_class_ids.item()]
+    return label
+
+def respond(message, history: list[tuple[str, str]], system_message, max_tokens, temperature, top_p):
     messages = [{"role": "system", "content": system_message}]
     for val in history:
         if val[0]:
@@ -32,17 +32,14 @@ def respond(
             messages.append({"role": "assistant", "content": val[1]})
     messages.append({"role": "user", "content": message})
     response = ""
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-        ):
+    for message in client.chat_completion(messages, max_tokens=max_tokens, stream=True, temperature=temperature, top_p=top_p): # Creo que lo importante para el modelo
         token = message.choices[0].delta.content
         response += token
         yield response
 
+def cambiar_pestaña():
+    return gr.update(visible=False), gr.update(visible=True)
+
 my_theme = gr.themes.Soft(
     primary_hue="emerald",
     secondary_hue="green",
@@ -60,24 +57,6 @@ my_theme = gr.themes.Soft(
     shadow_spread='*button_shadow_active'
     )
 
-def mostrar_pagina_1():
-    return gr.update(visible=False), gr.update(visible=True)
-
-def mostrar_pagina_2():
-    return gr.update(visible=False), gr.update(visible=True)
-
-def redirigir_a_pantalla_inicial():
-    return gr.update(visible=True), gr.update(visible=False)
-
-def transcribe(audio):
-    _, y = audio
-    y = y.astype(np.float32) # con torch.float32 da error
-    y /= np.max(np.abs(y))
-    results = pipe({"sampling_rate": SAMPLING_RATE, "raw": y})
-    top_result = results[0]  # Get the top result (most likely classification)
-    label = top_result["label"]  # Extract the label from the top result
-    return label
-
 with gr.Blocks(theme=my_theme) as demo:
     with gr.Column(visible=True, elem_id="pantalla-inicial") as pantalla_inicial:
         gr.HTML(
@@ -164,14 +143,13 @@ with gr.Blocks(theme=my_theme) as demo:
                 gr.Markdown("<h2>Predictor</h2>")
                 audio_input = gr.Audio(
                     min_length=1.0,
-                    # max_length=10.0,
                     format="wav",
-                    # type="numpy",
-                    label="Baby recorder"
-                    ),
+                    label="Baby recorder",
+                    type="filepath", # Para no usar numpy y preprocesar siempre igual
+                    )
                 classify_btn = gr.Button("¿Por qué llora?")
                 classification_output = gr.Textbox(label="Tu bebé llora por:")
-                classify_btn.click(transcribe, inputs=audio_input, outputs=classification_output)
+                classify_btn.click(predict, inputs=audio_input, outputs=classification_output)
             with gr.Column():
                 gr.Markdown("<h2>Assistant</h2>")
                 system_message = "You are a Chatbot specialized in baby health and care."
@@ -179,7 +157,7 @@ with gr.Blocks(theme=my_theme) as demo:
                 temperature = 0.7
                 top_p = 0.95
                 chatbot = gr.ChatInterface(
-                    respond,
+                    respond, # TODO: Cambiar para que argumentos estén aquí metidos
                     additional_inputs=[
                         gr.State(value=system_message),
                         gr.State(value=max_tokens),
@@ -188,13 +166,11 @@ with gr.Blocks(theme=my_theme) as demo:
                     ],
                 )
                 gr.Markdown("Este chatbot no sustituye a un profesional de la salud. Ante cualquier preocupación o duda, consulta con tu pediatra.")
-                boton_volver_inicio_1 = gr.Button("Volver a la pantalla inicial")
-        boton_volver_inicio_1.click(redirigir_a_pantalla_inicial, inputs=None, outputs=[pantalla_inicial, pagina_1])
+                boton_volver_inicio_1 = gr.Button("Volver a la pantalla inicial").click(cambiar_pestaña, outputs=[pagina_1, pantalla_inicial])
     with gr.Column(visible=False) as pagina_2:
         gr.Markdown("<h2>Monitor</h2>")
         gr.Markdown("Contenido de la Página 2")
-        boton_volver_inicio_2 = gr.Button("Volver a la pantalla inicial")
-        boton_volver_inicio_2.click(redirigir_a_pantalla_inicial, inputs=None, outputs=[pantalla_inicial, pagina_2])
-    boton_pagina_1.click(mostrar_pagina_1, inputs=None, outputs=[pantalla_inicial, pagina_1])
-    boton_pagina_2.click(mostrar_pagina_2, inputs=None, outputs=[pantalla_inicial, pagina_2])
+        boton_volver_inicio_2 = gr.Button("Volver a la pantalla inicial").click(cambiar_pestaña, outputs=[pagina_2, pantalla_inicial])
+    boton_pagina_1.click(cambiar_pestaña, outputs=[pantalla_inicial, pagina_1])
+    boton_pagina_2.click(cambiar_pestaña, outputs=[pantalla_inicial, pagina_2])
 demo.launch()