---
license: other
tags:
- image-tokenizer
- residual-vq
- text-rendering
- rda
---

# 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

```text
config.json
rda_model.pt
```

## Loading

```python
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")
```