Skip to main content

Fast inference and limited GPU memory

Fast Inference​

generate_fast.py runs accelerated causal inference with KV caching and processes video frames chunk by chunk:

Download models using huggingface-cli. (If you have not already downloaded lingbot-world-base-cam, please download it first.)

huggingface-cli download robbyant/lingbot-world-fast --local-dir ./lingbot-world-base-cam/lingbot_world_fast
  • LingBot-World-Fast — 480P, multi-GPU: 
    torchrun --nproc_per_node=8 generate_fast.py --task i2v-A14B --size 480*832 --ckpt_dir lingbot-world-base-cam --image examples/03/image.jpg --action_path examples/03 --dit_fsdp --t5_fsdp --ulysses_size 8 --frame_num 81 --prompt "A serene lakeside scene with a lone tree standing in calm water, surrounded by distant snow-capped mountains under a bright blue sky with drifting white clouds — gentle ripples reflect the tree and sky, creating a tranquil, meditative atmosphere."

You can also use the provided run_fast.sh script:

bash run_fast.sh <weights_dir> <frame_num>
# e.g. bash run_fast.sh lingbot-world-base-cam 201

Quantized Model for Limited GPU Resources​

For limited GPU memory, use the community-provided 4-bit quantized version of LingBot-World-Base (Cam). It reduces memory usage while preserving usable inference quality.

👉 Download link: https://huggingface.co/cahlen/lingbot-world-base-cam-nf4

⚠️ Note: This quantized model is intended for inference only. Minor degradation in visual fidelity and temporal consistency may occur compared to the full-precision model.

🎬 Demo Results​

⚡ Real-Time Interactive Demo Videos (Lingbot-World-Fast)​

These videos showcase Lingbot-World-Fast responding to user inputs and rendering results in real time.

🔍 Comparison Demo Videos (Lingbot-World-Base, Camera Pose Version)​

Camera parameters are estimated by ViPE from original videos downloaded from Genie3.