
SECourses TRELLIS - PyTorch 2.13.0 + CUDA 13.0 + Python 3.12

Fast Upload Download Data HuggingFace Tutorial : https://youtu.be/X5WVZ0NMaTg

SIMPLEPOD CHEAPER AND FASTER THAN RUNPOD

Now we fully support SimplePod as well please use this link to register : https://simplepod.ai/ref?user=secourses

SimplePod is faster and cheaper than RunPod and works exactly same

Please use this template on SimplePod : https://dash.simplepod.ai/account/explore/100/ref-secourses/

For permanent storage, generate it from Storage tab with any name and size you want and when selecting template with above link, click Edit and Use, select Persistence Volume and change mount point to /workspace

Up-to-date (22 June 2026) SimplePod tutorial (including how to use permanent storage) starting from 1:03:25 : https://youtu.be/9C_6qNKjgpA?si=ge1XvDomYQ7ssvMu&t=3805

RUNPOD

Please register for RunPod at: https://get.runpod.io/955rkuppqv4h

Up-to-date (22 June 2026) RunPod tutorial (including how to use permanent storage) starting from 31:01 : https://youtu.be/9C_6qNKjgpA?si=vEXM_L4V1L768f9f&t=1861

Select this template mandatory (based on official Runpod Pytorch template) : https://get.runpod.io/SECourses_CU13

Recommended GPU is RTX PRO 5090, but you can use any GPU. The more VRAM it is better

You can edit template and change Volume Disk size. it is by default 200 GB and should be sufficient without any changes

You can also use your permanent storage system - from filters select CUDA 12.8, 12.9, 13 and so on to get newer NVIDIA driver - only NVIDIA driver matter it must be 580+ for CUDA 13

CUDA 13 also needs a GPU of sm_75 or newer (RTX 20 series and up - Turing, Ampere, Ada, Hopper, Blackwell)

Upload zip file into workspace and extract there or extract in your system and upload all files into workspace

requirements_trellis.txt must sit next to RunPod_Trellis_Install.sh - it holds
every package version

Start a new terminal

To install run below command

cd /workspace
export HF_HOME="/workspace"
export UV_CONCURRENT_INSTALLS=4
chmod +x RunPod_Trellis_Install.sh
./RunPod_Trellis_Install.sh

The installer will install and download all models automatically

If your download fails for any reason, just run the install command again -
completed files are skipped and partial files resume where they stopped

Once the installation ended open a new terminal and copy paste below command to start the app. If you stop your pod and start again, run installer again to make sure all ready. Second install run should be always faster

unset LD_LIBRARY_PATH
cd /workspace/TRELLIS
git reset --hard
git pull
export HF_HOME="/workspace"
export TORCH_HOME="/workspace/torch"
export U2NET_HOME="/workspace/u2net"
export PYTHONWARNINGS=ignore
export HF_HUB_ENABLE_HF_TRANSFER=0
export CUDA_VISIBLE_DEVICES=0
source ./venv/bin/activate
python secourses_trellis.py --share --highvram

The TORCH_HOME and U2NET_HOME exports keep the DINOv2 and rembg weights on
your permanent volume - use the exact same values the installer used or they
get downloaded again on every pod start

To only re-download / resume models without touching the installation run:

cd /workspace
export HF_HOME="/workspace"
export TORCH_HOME="/workspace/torch"
export U2NET_HOME="/workspace/u2net"
source ./TRELLIS/venv/bin/activate
python DownloadModels.py

To kill previously running app without restarting the machine run below command

apt update
apt install -y psmisc
for p in $(seq 7860 8000); do fuser -k ${p}/tcp 2>/dev/null; done

If you get error during initial installation due to RunPod storage system errors, it corrupts your venv and you have to delete it and run installer again

To delete the venv run below command and run reinstall

cd /workspace/TRELLIS
rm -r venv
