[docs] Model cards #11112
[docs] Model cards #11112
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
stevhliu
force-pushed
the
video
branch
2 times, most recently
from
9cda27b to
b31ed8e
Compare
|
Looks good on first impression, I will review it in depth later today, wanted to raise #10301 with you as it will help simplify the examples (in combination with #11130 for the quantization cases). Also, it would be cool to have the examples be configurable/update with options, to demonstrate here's an artist's (4o) impression of what it could look like:
|
| |:---:|:---:| | ||
| | [`THUDM/CogVideoX-5b-I2V`](https://huggingface.co/THUDM/CogVideoX-5b-I2V) | torch.bfloat16 | | ||
| | [`THUDM/CogVideoX-1.5-5b-I2V`](https://huggingface.co/THUDM/CogVideoX-1.5-5b-I2V) | torch.bfloat16 | | ||
| [CogVideoX](https://huggingface.co/papers/2408.06072) is a large diffusion transformer model - available in 2B and 5B parameters - designed to generate longer and more consistent videos from text. This model uses a 3D causal variational autoencoder to more efficiently process video data by reducing sequence length (and associated training compute) and preventing flickering in generated videos. An "expert" transformer with adaptive LayerNorm improves alignment between text and video, and 3D full attention helps accurately capture motion and time in generated videos. |
There was a problem hiding this comment.
This is okay but I would perhaps tackle the removal of the abstract section in a separate PR. Also, this does add an additional overload of coming up with a description for the paper. I would like to avoid that for now.
There was a problem hiding this comment.
I think it'd be good to also tackle this now since for the new pipeline cards, we want to have a nice and complete example of what it should look like no?
Good point that adding a description of the paper adds additional overload, but I think its necessary, since we want to give users a version of the abstract that is more accessible (meaning using common everyday language) versus academic (inspired by @asomoza 's comment here)
There was a problem hiding this comment.
I am a bit spread thin on this one. So, I will go with what the team prefers.
|
Thanks @hlky, those PRs look to be super nice for user experience and I'll update the code examples once it's merged! The configurable example is also really neat and maybe we can make a Space out of it and embed it in the docs? I'll probably have to follow up on this one in a separate PR though 😅 |
|
@stevhliu sorry for the delay on my end. The changes look nice and I responded to some of the questions/comments you had. Perhaps after #11130, we could simplify the quantization examples a bit. @a-r-r-o-w do we want to touch any other video models in this PR? |
|
@sayakpaul, I simplified the examples with the new |
|
|
||
| ### Memory optimization | ||
| - CogVideoX supports LoRAs with [`~loaders.CogVideoXLoraLoaderMixin.load_lora_weights`]. |
There was a problem hiding this comment.
Do we need this separate note besides having the LoRA marker button at the top of the page?
There was a problem hiding this comment.
I think it'd be nice to have a easy copy/paste example for users who want to use this specific model, will fold under collapsible section as suggested. I also added a link to the LoRA marker button at the top :)
|
|
||
| Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`HunyuanVideoPipeline`] for inference with bitsandbytes. | ||
| Compilation is slow the first time but subsequent calls to the pipeline are faster. |
There was a problem hiding this comment.
For compilation, should we also refer the readers to our compilation guide?
There was a problem hiding this comment.
Added link to the compile section in fp16.md and will combine torch2.0.md with it in a separate PR as discussed!
🚧 WIP 🚧Based on our discussions about making it easier to run video models by including some minimal code optimized for memory and inference speed, this PR refactors the model card (starting with CogVideoX, but eventually expanding to other models as well) to reflect that. This provides users with easy copy/paste code they can run.
Parallel to this effort is to also improve the generic video generation guide.