Skip to content

quickstart.md

Latest commit

 

History

History
108 lines (85 loc) · 3.57 KB

quickstart.md

File metadata and controls

108 lines (85 loc) · 3.57 KB
Raw

Quickstart

This guide runs the Wan2.2-5B Stage0.5 153f recipe with released preencoded latents. This route does not require raw-WDS. For another model or training stage, finish the common setup below and continue with its backend guide.

1. Set up the Wan environment

Wan, LTX, and MiniMax-H3 use separate environments. Follow the Wan setup in Runtime environments, activate it, and install SolarWM:

export SOLAR_REPO=/path/to/SolarWM
cd "$SOLAR_REPO"
python -m pip install -e .
solarwm environment probe

2. Download weights and data

Choose local directories for the model, data, and outputs:

export SOLAR_MODEL_ROOT=/path/to/SolarWM-models
export SOLAR_DATA_HOME=/path/to/SolarWM-Data
export SOLAR_DATA_ROOT="$SOLAR_DATA_HOME/releases-v1"
export SOLAR_OUTPUT_ROOT=/path/to/outputs
mkdir -p "$SOLAR_MODEL_ROOT" "$SOLAR_DATA_HOME" "$SOLAR_OUTPUT_ROOT"

Download the Wan2.2-5B base model, its 81f initialization checkpoint, and the public data repository:

python -m pip install --upgrade huggingface_hub

hf download junchaoh-cs/SolarWM \
  --include "SolarWM-5B-base/**" \
  --include "SolarWM-5B-bid-stage0p5-81f/**" \
  --local-dir "$SOLAR_MODEL_ROOT"

hf download junchaoh-cs/SolarWM-Data \
  --repo-type dataset \
  --exclude "SolarWM-Data-Annotation/**" \
  --local-dir "$SOLAR_DATA_HOME"

Download wan22-ti2v5b-153f-480p-v1 and place the downloaded generation at:

$SOLAR_DATA_ROOT/latent-wds/wan22-ti2v5b-153f-480p-v1/

The main data repository supplies the matching recipe indexes. The latent generation supplies the training payload, so raw-WDS is not needed for this quickstart.

3. Check the configuration

Resolve the example with your local paths before starting training:

solarwm config resolve \
  --config configs/examples/wan22_ti2v_5b/train_stage0p5_fm_153f.yaml \
  --set model.base_path="$SOLAR_MODEL_ROOT/SolarWM-5B-base" \
  --set checkpoint.path="$SOLAR_MODEL_ROOT/SolarWM-5B-bid-stage0p5-81f/model.pt" \
  --set data.index_root="$SOLAR_DATA_ROOT" \
  --set data.transport.root="$SOLAR_DATA_ROOT" \
  --set runtime.validate_every=0 \
  --set runtime.output_dir="$SOLAR_OUTPUT_ROOT/wan5-stage0p5-153f"

4. Launch training

The following command runs the example on one eight-GPU node:

torchrun --standalone --nproc-per-node=8 \
  -m solarwm train \
  --config configs/examples/wan22_ti2v_5b/train_stage0p5_fm_153f.yaml \
  --set distributed.world_size=8 \
  --set train.global_batch_size=8 \
  --set model.base_path="$SOLAR_MODEL_ROOT/SolarWM-5B-base" \
  --set checkpoint.path="$SOLAR_MODEL_ROOT/SolarWM-5B-bid-stage0p5-81f/model.pt" \
  --set data.index_root="$SOLAR_DATA_ROOT" \
  --set data.transport.root="$SOLAR_DATA_ROOT" \
  --set runtime.validate_every=0 \
  --set runtime.output_dir="$SOLAR_OUTPUT_ROOT/wan5-stage0p5-153f"

Periodic video validation is disabled in this latent-only quickstart because the Wan validation recipe reads raw video. The output directory contains the resolved configuration, launch manifest, and checkpoints. Prepare raw-WDS only if you later need the full raw corpus, an online-encoding workflow, or raw-video validation and inference.

Next steps