CSU-JPG/RDA_llamagen

RDA LlamaGen Weights

This repository contains the Residual Decoder Adapter (RDA) / residual VQ weights for RDA tokenizer inference.

It does not include the base LlamaGen VQ checkpoint. Provide the base VQ checkpoint separately when loading the model.

Files

config.json
rda_model.pt

Loading

from tokenizer.tokenizer_image.rda_model import RDATokenizer

model = RDATokenizer.from_pretrained(
    "neversa/RDA_llamagen",
    vq_ckpt="/path/to/vq_ds16_t2i.pt",
).to("cuda")

outputs = model("/path/to/image.jpg")
outputs.prediction.save("prediction.png")