Skip to content

README.md

Latest commit

 

History

History
94 lines (81 loc) · 3.51 KB

README.md

File metadata and controls

94 lines (81 loc) · 3.51 KB
Raw

Runtime environments

Wan, LTX, and H3 require separate environments. The following matrix is the release-tested baseline; nearby compatible versions may work but are not part of the reproduced result.

Backend Python PyTorch / CUDA Required model stack
Wan 2.2 3.10 2.5.1 / 12.4 FlashAttention 2.8.3, Diffusers 0.38.0, Transformers 5.12.1
LTX-2.5 3.12 2.9.1 / 12.8 ltx-core, ltx-pipelines, and ltx-trainer 1.2.0; Transformers 5.14.1
MiniMax-H3 3.10 2.6.0 / 12.4 FlashAttention 2.8.3, PEFT 0.20.0, Transformers 5.12.1, Diffusers 0.40.0

Do not install these three stacks into one environment. Start from the matching CUDA image or build an equivalent environment, then install SolarWM without letting its optional helpers replace the already selected CUDA packages:

cd /path/to/SolarWM
python -m pip install -e .
solarwm environment probe

If the environment is managed by uv and intentionally has no pip module:

uv pip install --python /path/to/environment/bin/python -e /path/to/SolarWM
/path/to/environment/bin/python -m solarwm environment probe

Building without a prepared image

For Wan, create a Python 3.10 CUDA 12.4 environment and install PyTorch before FlashAttention. The following order avoids asking FlashAttention's build step to import a PyTorch installation that does not yet exist:

cd /path/to/SolarWM
python3.10 -m venv .venv-wan
source .venv-wan/bin/activate
python -m pip install --upgrade pip setuptools wheel
python -m pip install torch==2.5.1 \
  --index-url https://download.pytorch.org/whl/cu124
python -m pip install -e ".[train]" \
  diffusers==0.38.0 transformers==5.12.1 peft==0.20.0 \
  "ftfy>=6.2" "omegaconf>=2.3" "regex>=2024.0"
python -m pip install --no-build-isolation flash-attn==2.8.3
solarwm environment probe

For LTX, create a Python 3.12 environment, install the tested CUDA 12.8 PyTorch wheels first, then install the three official LTX packages from tag v1.2.0. Do not use the upstream uv sync command for this environment: that release has no lock file, and its default PyTorch index is not the CUDA 12.8 baseline used here.

cd /path/to/SolarWM
SOLAR_REPO="$PWD"
uv venv --python 3.12 .venv-ltx
uv pip install --python .venv-ltx/bin/python \
  torch==2.9.1 torchvision==0.24.1 torchaudio==2.9.1 \
  --index-url https://download.pytorch.org/whl/cu128
git clone https://github.com/Lightricks/LTX-2.git ../LTX-2
cd ../LTX-2
git checkout v1.2.0
uv pip install --python "$SOLAR_REPO/.venv-ltx/bin/python" \
  -e packages/ltx-core \
  -e packages/ltx-pipelines \
  -e packages/ltx-trainer \
  transformers==5.14.1 peft==0.20.0 safetensors==0.8.0
uv pip install --python "$SOLAR_REPO/.venv-ltx/bin/python" \
  -e "${SOLAR_REPO}[train,ltx]"
"$SOLAR_REPO/.venv-ltx/bin/python" -m solarwm environment probe

For H3, install PyTorch 2.6.0 and torchvision 0.21.0 first, then Diffusers 0.40.0, Transformers 5.12.1, PEFT 0.20.0, and FlashAttention 2.8.3. FlashAttention must be installed with build isolation disabled after PyTorch:

cd /path/to/SolarWM
python3.10 -m venv .venv-h3
source .venv-h3/bin/activate
python -m pip install --upgrade pip setuptools wheel
python -m pip install torch==2.6.0 torchvision==0.21.0 \
  --index-url https://download.pytorch.org/whl/cu124
python -m pip install \
  diffusers==0.40.0 \
  transformers==5.12.1 peft==0.20.0 \
  imageio==2.37.4 imageio-ffmpeg==0.6.0
python -m pip install -e ".[train]"
python -m pip install --no-build-isolation flash_attn==2.8.3
solarwm environment probe