Native UniCom SigLIP2-NaFlex vision, semantic compression/expansion, and flow interfaces.
Preserves 1152→64 semantic compression, 64→1152 expansion, vision aligner and latent/time input/output modules. The separate 23.8GB pixel decoder, FLUX VAE and language backbone are not included. IMPORTANT: the published checkpoint aligner requires 1216 input features, while the public direct vision helper emits 1152; this wrapper explicitly rejects implicit adaptation. No speculative concatenation is used. Compression and flow interfaces work independently and are tested.
Source and conversion
Model/weights: tencent/Unicom-Unified-Multimodal-Modeling-via-Compressed-Continuous-Semantic-Representations, fixed revision 78e43059d8358a045e64fca35698fda86c476154.
Native code: Tencent-Hunyuan/UniCom, fixed revision 55b9001a0fee623528676d995d8945fd6db7e23e.
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/unicom_vision_compression")
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
The License Terms of UniCom limit use to academic purposes and prohibit commercial or production use; see LICENSE.txt. The separately required FLUX decoder dependencies retain their own terms.
- Downloads last month
- 23