MiniGPT Casual + Yandere
Modelo GPT entrenado en conversaciones casuales y diálogos yandere.
Arquitectura
- Tokenizador: cl100k_base (tiktoken)
- Vocab size: 100,277
- Embedding: 256
- Capas: 6
- Heads: 8
- Block size: 128
Uso
import torch
from huggingface_hub import hf_hub_download
ruta = hf_hub_download("DarksitoBest/minigpt-casual-es", "pretrain.pt")
checkpoint = torch.load(ruta, map_location="cpu")
model = MiniGPT(n_heads=8, n_layers=6)
model.load_state_dict(checkpoint['model_state'])
- Downloads last month
- 10