mirage model for image understanding and extraction appears on HuggingFace
The model Cactus-Compute/mirage with the image-text-to-text task has been published on HuggingFace. The author describes it as a model for image understanding and structured information extraction. The publication date was 05.11.2026.
What it means
The largest weights file — mirage.safetensors — takes up 0.2 GB. The weights are published in safetensors format, and the license is apache-2.0. The model runs through the cactus-compute package, which downloads and caches the engine and weights on first use. There are no quantized builds on the platform yet.
The repository contains mirage.cact, mirage.safetensors and the model config. The author also points to the Python API documentation, the list of platforms supported by the runtime, and the source code. The developer gives no details on VRAM or CPU requirements.
For those following tools that can be run on your own hardware, this is another openly licensed model with a small footprint — more news in this area can be found in the CPU3D news feed.How the method works. The diagram is drawn based on this news note.