CogVideoX-2B

📄 中文阅读 | 🤗 Huggingface Space | 🌐 Github | 📜 arxiv

📍 Visit QingYing and API Platform to experience commercial video generation models.

Demo Show

A detailed wooden toy ship with intricately carved masts and sails is seen gliding smoothly over a plush, blue carpet that mimics the waves of the sea. The ship's hull is painted a rich brown, with tiny windows. The carpet, soft and textured, provides a perfect backdrop, resembling an oceanic expanse. Surrounding the ship are various other toys and children's items, hinting at a playful environment. The scene captures the innocence and imagination of childhood, with the toy ship's journey symbolizing endless adventures in a whimsical, indoor setting.

The camera follows behind a white vintage SUV with a black roof rack as it speeds up a steep dirt road surrounded by pine trees on a steep mountain slope, dust kicks up from it’s tires, the sunlight shines on the SUV as it speeds along the dirt road, casting a warm glow over the scene. The dirt road curves gently into the distance, with no other cars or vehicles in sight. The trees on either side of the road are redwoods, with patches of greenery scattered throughout. The car is seen from the rear following the curve with ease, making it seem as if it is on a rugged drive through the rugged terrain. The dirt road itself is surrounded by steep hills and mountains, with a clear blue sky above with wispy clouds.

A street artist, clad in a worn-out denim jacket and a colorful bandana, stands before a vast concrete wall in the heart, holding a can of spray paint, spray-painting a colorful bird on a mottled wall.

In the haunting backdrop of a war-torn city, where ruins and crumbled walls tell a story of devastation, a poignant close-up frames a young girl. Her face is smudged with ash, a silent testament to the chaos around her. Her eyes glistening with a mix of sorrow and resilience, capturing the raw emotion of a world that has lost its innocence to the ravages of conflict.

Model Introduction

CogVideoX is an open-source version of the video generation model originating from QingYing. The table below displays the list of video generation models we currently offer, along with their foundational information.

Model Name	CogVideoX-2B (This Repository)	CogVideoX-5B
Model Description	Entry-level model, balancing compatibility. Low cost for running and secondary development.	Larger model with higher video generation quality and better visual effects.
Inference Precision	FP16* (Recommended), BF16, FP32, FP8*, INT8, no support for INT4	BF16 (Recommended), FP16, FP32, FP8*, INT8, no support for INT4
Single GPU VRAM Consumption	SAT FP16: 18GB diffusers FP16: starting from 4GB* diffusers INT8(torchao): starting from 3.6GB*	SAT BF16: 26GB diffusers BF16: starting from 5GB* diffusers INT8(torchao): starting from 4.4GB*
Multi-GPU Inference VRAM Consumption	FP16: 10GB* using diffusers	BF16: 15GB* using diffusers
Inference Speed (Step = 50, FP/BF16)	Single A100: ~90 seconds Single H100: ~45 seconds	Single A100: ~180 seconds Single H100: ~90 seconds
Fine-tuning Precision	FP16	BF16
Fine-tuning VRAM Consumption (per GPU)	47 GB (bs=1, LORA) 61 GB (bs=2, LORA) 62GB (bs=1, SFT)	63 GB (bs=1, LORA) 80 GB (bs=2, LORA) 75GB (bs=1, SFT)
Prompt Language	English*
Prompt Length Limit	226 Tokens
Video Length	6 Seconds
Frame Rate	8 Frames per Second
Video Resolution	720 x 480, no support for other resolutions (including fine-tuning)
Positional Encoding	3d_sincos_pos_embed	3d_rope_pos_embed

Data Explanation

• When testing using the diffusers library, all optimizations provided by the diffusers library were enabled. This solution has not been tested for actual VRAM/memory usage on devices other than NVIDIA A100 / H100. Generally, this solution can be adapted to all devices with NVIDIA Ampere architecture and above. If the optimizations are disabled, VRAM usage will increase significantly, with peak VRAM usage being about 3 times higher than the table shows. However, speed will increase by 3-4 times. You can selectively disable some optimizations, including:

pipe.enable_model_cpu_offload()
pipe.enable_sequential_cpu_offload()
pipe.vae.enable_slicing()
pipe.vae.enable_tiling()

• When performing multi-GPU inference, the enable_model_cpu_offload() optimization needs to be disabled.
• Using INT8 models will reduce inference speed. This is to ensure that GPUs with lower VRAM can perform inference normally while maintaining minimal video quality loss, though inference speed will decrease significantly.
• The 2B model is trained with FP16 precision, and the 5B model is trained with BF16 precision. We recommend using the precision the model was trained with for inference.
• PytorchAO and Optimum-quanto can be used to quantize the text encoder, Transformer, and VAE modules to reduce CogVideoX's memory requirements. This makes it possible to run the model on a free T4 Colab or GPUs with smaller VRAM! It is also worth noting that TorchAO quantization is fully compatible with torch.compile, which can significantly improve inference speed. FP8 precision must be used on devices with NVIDIA H100 or above, which requires installing the torch, torchao, diffusers, and accelerate Python packages from source. CUDA 12.4 is recommended.
• The inference speed test also used the above VRAM optimization scheme. Without VRAM optimization, inference speed increases by about 10%. Only the diffusers version of the model supports quantization.
• The model only supports English input; other languages can be translated into English during refinement by a large model.

Note

• Using SAT for inference and fine-tuning of SAT version models. Feel free to visit our GitHub for more information.

Quick Start 🤗

This model supports deployment using the huggingface diffusers library. You can deploy it by following these steps.

We recommend that you visit our GitHub and check out the relevant prompt optimizations and conversions to get a better experience.

1. Install the required dependencies

# diffusers>=0.30.1
# transformers>=0.44.0
# accelerate>=0.33.0 (suggest install from source)
# imageio-ffmpeg>=0.5.1
pip install --upgrade transformers accelerate diffusers imageio-ffmpeg 

2. Run the code

import torch
from diffusers import CogVideoXPipeline
from diffusers.utils import export_to_video

prompt = "A panda, dressed in a small, red jacket and a tiny hat, sits on a wooden stool in a serene bamboo forest. The panda's fluffy paws strum a miniature acoustic guitar, producing soft, melodic tunes. Nearby, a few other pandas gather, watching curiously and some clapping in rhythm. Sunlight filters through the tall bamboo, casting a gentle glow on the scene. The panda's face is expressive, showing concentration and joy as it plays. The background includes a small, flowing stream and vibrant green foliage, enhancing the peaceful and magical atmosphere of this unique musical performance."

pipe = CogVideoXPipeline.from_pretrained(
    "THUDM/CogVideoX-2b",
    torch_dtype=torch.float16
)

pipe.enable_model_cpu_offload()
pipe.enable_sequential_cpu_offload()
pipe.vae.enable_slicing()
pipe.vae.enable_tiling()
video = pipe(
    prompt=prompt,
    num_videos_per_prompt=1,
    num_inference_steps=50,
    num_frames=49,
    guidance_scale=6,
    generator=torch.Generator(device="cuda").manual_seed(42),
).frames[0]

export_to_video(video, "output.mp4", fps=8)

Quantized Inference

PytorchAO and Optimum-quanto can be used to quantize the Text Encoder, Transformer and VAE modules to lower the memory requirement of CogVideoX. This makes it possible to run the model on free-tier T4 Colab or smaller VRAM GPUs as well! It is also worth noting that TorchAO quantization is fully compatible with torch.compile, which allows for much faster inference speed.

# To get started, PytorchAO needs to be installed from the GitHub source and PyTorch Nightly.
# Source and nightly installation is only required until next release.

import torch
from diffusers import AutoencoderKLCogVideoX, CogVideoXTransformer3DModel, CogVideoXPipeline
from diffusers.utils import export_to_video
+ from transformers import T5EncoderModel
+ from torchao.quantization import quantize_, int8_weight_only, int8_dynamic_activation_int8_weight

+ quantization = int8_weight_only

+ text_encoder = T5EncoderModel.from_pretrained("THUDM/CogVideoX-5b", subfolder="text_encoder", torch_dtype=torch.bfloat16)
+ quantize_(text_encoder, quantization())

+ transformer = CogVideoXTransformer3DModel.from_pretrained("THUDM/CogVideoX-5b", subfolder="transformer", torch_dtype=torch.bfloat16)
+ quantize_(transformer, quantization())

+ vae = AutoencoderKLCogVideoX.from_pretrained("THUDM/CogVideoX-2b", subfolder="vae", torch_dtype=torch.bfloat16)
+ quantize_(vae, quantization())

# Create pipeline and run inference
pipe = CogVideoXPipeline.from_pretrained(
    "THUDM/CogVideoX-2b",
+    text_encoder=text_encoder,
+    transformer=transformer,
+    vae=vae,
    torch_dtype=torch.bfloat16,
)
pipe.enable_model_cpu_offload()
pipe.vae.enable_tiling()

prompt = "A panda, dressed in a small, red jacket and a tiny hat, sits on a wooden stool in a serene bamboo forest. The panda's fluffy paws strum a miniature acoustic guitar, producing soft, melodic tunes. Nearby, a few other pandas gather, watching curiously and some clapping in rhythm. Sunlight filters through the tall bamboo, casting a gentle glow on the scene. The panda's face is expressive, showing concentration and joy as it plays. The background includes a small, flowing stream and vibrant green foliage, enhancing the peaceful and magical atmosphere of this unique musical performance."

video = pipe(
    prompt=prompt,
    num_videos_per_prompt=1,
    num_inference_steps=50,
    num_frames=49,
    guidance_scale=6,
    generator=torch.Generator(device="cuda").manual_seed(42),
).frames[0]

export_to_video(video, "output.mp4", fps=8)

Additionally, the models can be serialized and stored in a quantized datatype to save disk space when using PytorchAO. Find examples and benchmarks at these links:

• torchao
• quanto

Explore the Model

Welcome to our github, where you will find:

1. More detailed technical details and code explanation.
2. Optimization and conversion of prompt words.
3. Reasoning and fine-tuning of SAT version models, and even pre-release.
4. Project update log dynamics, more interactive opportunities.
5. CogVideoX toolchain to help you better use the model.
6. INT8 model inference code support.

Model License

The CogVideoX-2B model (including its corresponding Transformers module and VAE module) is released under the Apache 2.0 License.

The CogVideoX-5B model (Transformers module) is released under the CogVideoX LICENSE.

Citation

@article{yang2024cogvideox,
  title={CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer},
  author={Yang, Zhuoyi and Teng, Jiayan and Zheng, Wendi and Ding, Ming and Huang, Shiyu and Xu, Jiazheng and Yang, Yuanming and Hong, Wenyi and Zhang, Xiaohan and Feng, Guanyu and others},
  journal={arXiv preprint arXiv:2408.06072},
  year={2024}
}