app.py

import os
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"  # CPU

import gradio as gr
import tensorflow as tf
from huggingface_hub import hf_hub_download
from transformers import AutoTokenizer, TFAutoModel


@tf.keras.utils.register_keras_serializable()
class DistilBertLayer(tf.keras.layers.Layer):
    def __init__(self, model_name="vinai/bertweet-base", **kwargs):
        super().__init__(**kwargs)
        self.model_name = model_name
        self.bert = TFAutoModel.from_pretrained(model_name, from_pt=True)

    def call(self, inputs):
        input_ids, attention_mask = inputs
        outputs = self.bert(
            input_ids=input_ids,
            attention_mask=attention_mask,
            training=False
        )
        return outputs.last_hidden_state

    def get_config(self):
        config = super().get_config()
        config.update({"model_name": self.model_name})
        return config


# 1) Repo donde subiste el .keras (MODELS, no Spaces)
MODEL_REPO = "tomy07417/disaster-tweets-bertweet-gru"  # <-- CAMBIÁ ESTO
MODEL_FILE = "bertweet_gru_model.keras"                # <-- nombre exacto en el repo

# 2) Descarga con cache (no lo baja cada vez)
model_path = hf_hub_download(
    repo_id=MODEL_REPO,
    filename=MODEL_FILE,
    repo_type="model"
)

# 3) Cargar el modelo desde el path descargado
model = tf.keras.models.load_model(
    model_path,
    custom_objects={"DistilBertLayer": DistilBertLayer},
    compile=False
)

tokenizer = AutoTokenizer.from_pretrained("vinai/bertweet-base")


def predict(text):
    inputs = tokenizer(
        [text],
        max_length=50,
        truncation=True,
        padding="max_length",
        return_tensors="tf"
    )

    input_ids = inputs["input_ids"]
    attention_mask = inputs["attention_mask"]

    # si tu salida es (1,) sigmoid:
    prob = model.predict([input_ids, attention_mask])[0][0]
    pred = bool(prob > 0.5)

    return {"prob": float(prob), "pred": pred}


demo = gr.Interface(
    fn=predict,
    inputs=gr.Textbox(lines=3, label="Tweet"),
    outputs=gr.JSON(label="Result"),
    title="Tweet classifier",
    description="Paste a tweet in English"
)

if __name__ == "__main__":
    # En Spaces NO uses share=True
    demo.launch(ssr_mode=False)