This model requires you to accept the Apache 2.0 license terms before download.
Description
Text-to-video diffusion transformer, 2B parameters, 720x480 clips with classifier-free guidance.
Tags
At a glance
| Kind | video generation model |
|---|---|
| Precision | float16 |
| Container size | 12.9 GiB |
| Weights inside | 12.8 GiB |
| Components | 3 |
| Licence | Apache 2.0 |
| Origin | THUDM/CogVideoX-2b @ 1137dacfc2c9 |
What is inside the container
A .nbx container packs every part of CogVideoX-2b — 3 components with weights, graphs and runtime settings — so one file is all the runtime needs.
| Component | Blocks | Layers | Weights |
|---|---|---|---|
| transformer | 30 | 30 | 3.2 GiB |
| vae | 4 | — | 822 MiB |
| text_encoder | 24 | 24 | 8.9 GiB |
Default settings
The values the container ships with. They apply when you do not pass your own.
- Frames
- 49
- Frames per second
- 8
- Inference steps
- 50
- Guidance scale
- 6
Memory needed
The weights of CogVideoX-2b take 12.8 GiB in float16. 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 12.9 GiB (13,827,390,501 bytes).
$ sha256sum CogVideoX-2b.nbx
d97a4f689032ee0d4d9596ae367c60c799fb7597efe716ee40586e54a2b7cd52
Questions
What is CogVideoX-2b?
A video generation model by zai-org, distributed as a float16 .nbx container and licensed Apache 2.0.
How big is the download?
12.9 GiB, of which 12.8 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 zai-org/CogVideoX-2b, then neurobrix serve --model CogVideoX-2b.
CLI Usage
$ neurobrix import zai-org/CogVideoX-2b
$ neurobrix serve --model CogVideoX-2b
$ neurobrix run --prompt "your prompt here"
$ neurobrix stop