TRELLIS Colab Setup

---

Trellis colab notebook. jupyter. whatever. working setup! This README details the environment setup, package versions, and steps used to run the Microsoft TRELLIS project, along with the mip-splatting submodules, FlashAttention (2.5.8), spconv, and Kaolin (0.17.0). Everything is pinned to CUDA 11.8 and PyTorch 2.3.0 for reproducibility.

If you are looking for just the Google Colab - go here: Open In Colab

Table of Contents

1. Runtime Requirements
2. Dependencies & Versions
3. Steps to Reproduce in Google Colab
   1. Install CUDA 11.8
   2. Create Conda Environment (trellis)
   3. Install System Dev Tools
   4. Clone & Setup TRELLIS
   5. Clone & Setup mip-splatting
   6. Launching TRELLIS with Ngrok
4. References & Links

---

Runtime Requirements

• NVIDIA GPU with CUDA Compute Capability >= 7.0 (recommended)
• Google Colab GPU Runtime (or equivalent local environment)
• Ubuntu 22.04 or a similar Linux environment for best compatibility
• Miniconda/Conda environment management

When using Colab, be sure to enable a GPU runtime:

Runtime > Change runtime type > Hardware accelerator: GPU.

---

Dependencies & Versions

Below is a summary of major packages, pinned versions, and the sources we use:

1. Operating System

   • Ubuntu 22.04 (default Google Colab base image)

2. CUDA

   • CUDA 11.8 (installed from NVIDIA’s Ubuntu 22.04 repo)
   • GPG key and repo from:
     [https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/](https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/)

3. Python

   • Python 3.11 (via Miniconda)

4. Conda Environment

   • Environment name: trellis

5. PyTorch & Related

   • PyTorch 2.3.0
   • TorchVision 0.18.0
   • TorchAudio 2.3.0
   • pytorch-cuda=11.8

6. FlashAttention

   • flash-attn==2.5.8 (cu118, torch2.3, cxx11abiFALSE build) from
     https://github.com/Dao-AILab/flash-attention

7. spconv

   • spconv-cu118 (PyPI)

8. Kaolin

   • kaolin==0.17.0, installed from
     https://nvidia-kaolin.s3.us-east-2.amazonaws.com/torch-2.3.0_cu118.html

9. Pillow

   • pillow<11.0 (forced reinstall to avoid compatibility issues)

10. Gradio

    • gradio==4.44.1
    • gradio_litmodel3d==0.0.1

11. System Tools

    • build-essential, cmake, ninja-build, nvidia-cuda-toolkit
    • Python build libraries: pip, setuptools, wheel, ninja, Cython

12. Microsoft TRELLIS

    • Cloned from https://github.com/microsoft/TRELLIS with submodules
    • Submodule builds: flash-attn, spconv, mipgaussian, nvdiffrast, kaolin
    • Patched app.py to use demo.launch(share=True)

13. mip-splatting

    • Cloned from https://github.com/autonomousvision/mip-splatting
    • Installs the diff_gaussian_rasterization submodule

14. Ngrok

    • Used to expose the local Gradio app externally
    • Make sure you have an ngrok auth token

---

Steps to Reproduce in Google Colab

(or just view the kaiser-colab file for full cells)

1. Install CUDA 11.8

1. Update apt and install gnupg
2. Download and install cuda-keyring_1.1-1_all.deb from NVIDIA
3. apt-get update
4. Install cuda-11-8
5. (Optional) Check nvcc --version

# Example snippet:
sudo apt-get update -y
sudo apt-get install -y gnupg
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt-get update -y
sudo apt-get install -y cuda-11-8
export PATH="/usr/local/cuda-11.8/bin:$PATH"
nvcc --version

2. Create Conda Environment (trellis)

1. Install Miniconda inside Colab using condacolab
2. Activate conda environment in a Bash cell
3. Create trellis with Python 3.11
4. Install PyTorch 2.3.0 + CUDA 11.8, flash-attn, spconv, kaolin, etc.

# Example snippet:
pip install -q condacolab
python -c "import condacolab;condacolab.install_miniconda()"

# Then in a bash cell:
source /usr/local/etc/profile.d/conda.sh
conda create -y -n trellis python=3.11
conda activate trellis
conda install pytorch==2.3.0 torchvision==0.18.0 torchaudio==2.3.0 pytorch-cuda=11.8 -c pytorch -c nvidia
pip install https://github.com/Dao-AILab/flash-attention/releases/download/v2.5.8/flash_attn-2.5.8+cu118torch2.3cxx11abiFALSE-cp311-cp311-linux_x86_64.whl
pip install spconv-cu118
pip install kaolin==0.17.0 -f https://nvidia-kaolin.s3.us-east-2.amazonaws.com/torch-2.3.0_cu118.html
pip install --force-reinstall "pillow<11.0"
pip install gradio==4.44.1 gradio_litmodel3d==0.0.1

3. Install System Dev Tools

1. Install build-essential, cmake, ninja-build, nvidia-cuda-toolkit
2. Activate trellis again
3. Upgrade Python build libraries

apt-get update -y
apt-get install -y build-essential cmake ninja-build nvidia-cuda-toolkit
source /usr/local/etc/profile.d/conda.sh
conda activate trellis
pip install --upgrade pip setuptools wheel ninja Cython

4. Clone & Setup TRELLIS

1. Remove any previous /content/TRELLIS
2. git clone --recurse-submodules https://github.com/microsoft/TRELLIS.git /content/TRELLIS
3. cd /content/TRELLIS && git submodule update --init --recursive
4. Create assets/example_image so app.py doesn’t crash
5. Run setup.sh --basic --flash-attn --spconv --mipgaussian --nvdiffrast --kaolin --demo
6. Patch app.py to set demo.launch(share=True)

rm -rf /content/TRELLIS
git clone --recurse-submodules https://github.com/microsoft/TRELLIS.git /content/TRELLIS
cd /content/TRELLIS
git submodule update --init --recursive
mkdir -p /content/TRELLIS/assets/example_image
bash ./setup.sh --basic --flash-attn --spconv --mipgaussian --nvdiffrast --kaolin --demo
sed -i 's/demo.launch()$/demo.launch(share=True)/' app.py

5. Clone & Setup mip-splatting

1. git clone --recurse-submodules https://github.com/autonomousvision/mip-splatting.git
2. cd mip-splatting && git submodule update --init --recursive
3. Install system libs: libgl1-mesa-dev, etc.
4. Install diff_gaussian_rasterization submodule

cd /content
git clone --recurse-submodules https://github.com/autonomousvision/mip-splatting.git
cd mip-splatting
git submodule update --init --recursive
apt-get install -y libgl1-mesa-dev
cd submodules/diff-gaussian-rasterization
pip install . --no-build-isolation -v

6. Launching TRELLIS with Ngrok

1. Make sure ngrok is installed in your environment or is available on Colab
2. Set your ngrok auth token (replace YOUR_NGROK_AUTH_TOKEN below)
3. Expose port 7860
4. Run app.py in the trellis environment

import subprocess, time, json

NGROK_AUTH = "YOUR_NGROK_AUTH_TOKEN"

subprocess.run(["ngrok", "config", "add-authtoken", NGROK_AUTH], check=True)
proc_ngrok = subprocess.Popen(["ngrok", "http", "7860"])

time.sleep(4)
tunnels_json = subprocess.check_output(["curl", "-s", "127.0.0.1:4040/api/tunnels"])
public_url = json.loads(tunnels_json)["tunnels"][0]["public_url"]
print("NGROK URL =>", public_url)

p = subprocess.Popen(
    ["conda", "run", "-n", "trellis", "python", "-u", "app.py"],
    cwd="/content/TRELLIS",
    stdout=subprocess.PIPE,
    stderr=subprocess.STDOUT
)

try:
    for line in iter(p.stdout.readline, b""):
        if not line:
            break
        print(line.decode(), end="")
except KeyboardInterrupt:
    print("[INFO] Interrupted by user.")
    p.kill()
finally:
    p.wait()

You will see logs from the Gradio/TRELLIS app in real-time. Click the NGROK URL to open the Gradio interface in your browser.

---

References & Links

1. TRELLIS (Microsoft)

   • GitHub: https://github.com/microsoft/TRELLIS

2. mip-splatting (autonomousvision)

   • GitHub: https://github.com/autonomousvision/mip-splatting

3. FlashAttention

   • https://github.com/Dao-AILab/flash-attention

4. Kaolin (NVIDIA)

   • https://github.com/NVIDIAGameWorks/kaolin/
   • Wheels: https://nvidia-kaolin.s3.us-east-2.amazonaws.com/torch-2.3.0_cu118.html

5. Ngrok

   • https://ngrok.com/
   • After you sign up, get your token from the dashboard.

6. condacolab

   • https://github.com/conda-incubator/condacolab

---

Contributing

Feel free to open issues or pull requests if you encounter any build problems or want to add instructions for other OS distributions.

Enjoy exploring TRELLIS, Mip-Splatting, and advanced PyTorch-based 3D frameworks!