This model requires you to accept the MIT license terms before download.
Description
Janus-Pro-7B unified multimodal (image generation). Validated engine surface: PyTorch branch (compiled + sequential, byte-identical); image-autoregressive on the Triton engine is pending and refused with a clear message until then.
At a glance
| Kind | multimodal model |
|---|---|
| Precision | bfloat16 |
| Container size | 13.8 GiB |
| Weights inside | 13.8 GiB |
| Components | 7 |
| Licence | MIT |
What is inside the container
A .nbx container packs every part of Janus-Pro-7B — 7 components with weights, graphs and runtime settings — so one file is all the runtime needs.
| Component | Blocks | Layers | Weights |
|---|---|---|---|
| aligner | 0 | — | 40.0 MiB |
| gen_aligner | 0 | — | 32.1 MiB |
| gen_embed | 0 | — | 0.3 MiB |
| gen_head | 0 | — | 160 MiB |
| gen_vision_model | 0 | — | 137 MiB |
| language_model | 0 | — | 12.9 GiB |
| vision_model | 0 | — | 603 MiB |
Default settings
The values the container ships with. They apply when you do not pass your own.
- Height (px)
- 384
- Width (px)
- 384
- Guidance scale
- 5
- Temperature
- 1
- Top-p
- 1
- Top-k
- 0
Memory needed
The weights of Janus-Pro-7B take 13.8 GiB in bfloat16. That is the floor for the memory the runtime has to place, spread over the devices it finds; activations come on top and grow with the size and length you ask for. A measured peak per device class is not published for this model yet.
Verify what you downloaded
Every download carries its SHA-256, and the file endpoint sends it in the X-NBX-SHA256 header. The container is 13.8 GiB (14,842,924,358 bytes).
$ sha256sum Janus-Pro-7B.nbx
868f975f2ddb01d5552a0958c1a01fcd9b8d9dda834c5b76c57e033f8bccb3e2
Questions
What is Janus-Pro-7B?
A multimodal model by deepseek-ai, distributed as a bfloat16 .nbx container and licensed MIT.
How big is the download?
13.8 GiB, of which 13.8 GiB are weights.
How do I run it?
Install the runtime from the download page, then use the commands under “CLI Usage” below: neurobrix import deepseek-ai/Janus-Pro-7B, then neurobrix serve --model Janus-Pro-7B.
CLI Usage
$ neurobrix import deepseek-ai/Janus-Pro-7B
$ neurobrix serve --model Janus-Pro-7B
$ neurobrix run --prompt "your prompt here"
$ neurobrix stop