No output video

[tpmccallum]

I am using the following commands to execute text2world:

tpmccallum:~$ docker run -m 16g --gpus all --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 -it -v $(pwd):/workspace cosmos bash

root@928d748a0913:/workspace# PROMPT="A sleek, humanoid robot stands in a vast warehouse filled with neatly stacked cardboard boxes on industrial shelves."
root@928d748a0913:/workspace# export PYTHONPATH=$(pwd)
root@928d748a0913:/workspace# python -c "import sys; print(sys.path)"
root@928d748a0913:/workspace# cd Cosmos

root@928d748a0913:/workspace# python cosmos1/models/diffusion/inference/text2world.py --video_save_folder output --checkpoint_dir /workspace/checkpoints --diffusion_transformer_dir Cosmos-1.0-Diffusion-7B-Text2World --prompt '$PROMPT' --offload_prompt_upsampler --video_save_name Cosmos-1.0-Diffusion-7B-Text2World

The output does not show any errors, the processes finish but no video is saved to the output directory.

[tpmccallum]

The text2world script does not fail. But these warnings might be why no video is being generated:

  warnings.warn(
Import succeeded.
/usr/local/lib/python3.10/dist-packages/torch/cuda/__init__.py:210: UserWarning: 
    Found GPU0 Quadro P2200 which is of cuda capability 6.1.
    PyTorch no longer supports this GPU because it is too old.
    The minimum cuda capability supported by this library is 7.0.
    
  warnings.warn(
/usr/local/lib/python3.10/dist-packages/torch/cuda/__init__.py:235: UserWarning: 
Quadro P2200 with CUDA capability sm_61 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_70 sm_72 sm_75 sm_80 sm_86 sm_87 sm_90 compute_90.
If you want to use the Quadro P2200 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

I am running on an Ubuntu machine that is a few years old.

[ymcki]

Well, Cosmos only supports Ampere, Hopper, Ada Lovelace and Blackwell architecture. So you are out of luck unless you have these cards.

[tpmccallum]

Thanks for the information @ymcki
In that case, I have created a PR that lets the user know about the architecture support at the outset.
#57

[sophiahhuang]

Thanks for your interest in Cosmos. Cosmos is currently only supported on Ampere and above, please see (Ampere and above)

[tpmccallum]

Hi @sophiahhuang
Can you approve and merge the PR that I suggested #57

At least then, users will know before they start using Cosmos.

[sophiahhuang]

@tpmccallum We are conducting review and will update shortly.