24 lines
938 B
Bash
Executable File
24 lines
938 B
Bash
Executable File
#!/bin/bash
|
|
# Download the preprocessed MixKit finetune dataset used for the QAD 5090 recipe.
|
|
#
|
|
# This is the MixKit subset already VAE-encoded (Wan2.1-T2V-1.3B) and text-embedded
|
|
# into Parquet shards, so it can be fed straight to training with no further
|
|
# preprocessing. To build the Parquet from raw videos yourself, see
|
|
# examples/training/finetune/wan_t2v_1.3B/mixkit/README.md.
|
|
#
|
|
# Usage (run from the repo root):
|
|
# bash examples/datasets/mixkit/download_dataset.sh [DATA_ROOT]
|
|
set -euo pipefail
|
|
|
|
DATA_ROOT=${1:-data/HD-Mixkit-Finetune-Wan}
|
|
|
|
python scripts/huggingface/download_hf.py \
|
|
--repo_id "weizhou03/HD-Mixkit-Finetune-Wan" \
|
|
--local_dir "${DATA_ROOT}" \
|
|
--repo_type "dataset"
|
|
|
|
echo "Done."
|
|
echo " Train data: ${DATA_ROOT}/combined_parquet_dataset"
|
|
echo " Validation data: ${DATA_ROOT}/validation_parquet_dataset"
|
|
echo "Point your training script's data path at the combined_parquet_dataset directory."
|