Native multi-codebook input/output interfaces from the NVIDIA InternLumina-U2 checkpoint.

Contains eight input embedding tables (5010 rows each), concatenation projection, eight 4096-class output heads, and the six-layer native MultiCodebookARHead. AToken codes are 0..4095; input embeddings also include reserved/mask IDs. The 16.9B shared diffusion backbone and Apple AToken weights are separate dependencies. This interface bundle is not a standalone image generator.

Source and conversion

Model/weights: internlm/InternLumina-U2, fixed revision 270b4a23a5c4f480250d2eb5390aa86d8ca9e70a. Native code: InternLM/InternLumina-U2, fixed revision e91ee81a56c6b24a1b442828e04699798a5fb524.

All selected tensor names, shapes, dtypes and payload bytes are preserved, except the explicit IBQ EMA shadow-name mapping. See extraction.json, provenance.json and tensor/conversion mappings. This repository contains component weights and a local native loader, not an AutoModel.from_pretrained checkpoint.

Load

Install requirements.txt in a compatible environment, then:

from huggingface_hub import snapshot_download
import sys
import torch
root = snapshot_download("gatilin/internlumina_u2_visual_io")
sys.path.insert(0, root)
from component import load_model
model = load_model(root, dtype=torch.float32, device="cpu")

The model is returned in evaluation mode. Run inference under torch.inference_mode(). Inputs are native tensors/patches, not arbitrary PIL images; inspect component.py and the included validation script for the exact layout. CPU float32 was tested; CUDA/MPS and full end-to-end benchmarks were not tested.

Validation

See validation.json for the executed component tests, shapes and limitations. Source payload equality and native strict loading are required before release. Component tests do not establish full-model generation or understanding quality. Original stochastic decoding settings and normalization conventions must be retained.

License and attribution

Original upstream notices and license files are retained. Reused third-party components keep their original terms. This extraction does not claim authorship of the underlying trained model.