# precalcular_modelos_b16.py
import torch
import open_clip
import pandas as pd

# Solo marca + modelo
df = pd.read_excel("modelos.xlsx")
textos = (df["Marca"] + " " + df["Modelo"]).tolist()

MODEL_NAME = "ViT-B-16"
PRETRAINED = "openai"

model, _, _ = open_clip.create_model_and_transforms(MODEL_NAME, pretrained=PRETRAINED)
tokenizer = open_clip.get_tokenizer(MODEL_NAME)

device = "cuda" if torch.cuda.is_available() else "cpu"
model = model.to(device)

with torch.no_grad():
    text_inputs = tokenizer(textos).to(device)   # tensor en GPU o CPU
    text_features = model.encode_text(text_inputs)
    text_features /= text_features.norm(dim=-1, keepdim=True)

torch.save({"embeddings": text_features.cpu(), "labels": textos}, "text_embeddings_modelos_b16.pt")
print("Embeddings de modelos guardados en 'text_embeddings_modelos_b16.pt'")