docs/sampling_dataset_dreamdojo.md

# Create batch input for dreamgen
## Dataset: PhysicalAI-Robotics-GR00T-GR1/gr1

### Dense 
export PYTHONPATH="./models/dreamgen"
export CUDA_VISIBLE_DEVICES=0 
export NUM_GPUS=1
python -m scripts.prepare_batch_input_json_csv \
  --dataset_path ./datasets/PhysicalAI-Robotics-GR00T-GR1/gr1 \
  --metadata_path ./datasets/PhysicalAI-Robotics-GR00T-GR1/metadata.csv \
  --save_path ./sampling_dataset/PhysicalAI-Robotics-GR00T-GR1/gr1/dense \
  --output_path ./sampling_dataset/PhysicalAI-Robotics-GR00T-GR1/gr1/dense/batch_input.json

source /pfss/mlde/workspaces/mlde_wsp_IAS_SAMMerge/VLA/doanh/video_world/video_gen_physics/models/DreamDojo/.venv/bin/activate
export PYTHONPATH="./models/dreamgen"
export CUDA_VISIBLE_DEVICES=0
export NUM_GPUS=1
torchrun -m models.dreamgen.examples.video2world_gr00t \
  --model_size 14B \
  --gr00t_variant gr1 \
  --batch_input_json ./sampling_dataset/PhysicalAI-Robotics-GR00T-GR1/gr1/dense/batch_input.json \
  --disable_guardrail \
  --num_gpus ${NUM_GPUS}

### pisa
export PYTHONPATH="./models/dreamgen"
export CUDA_VISIBLE_DEVICES=0 
export NUM_GPUS=1
python -m scripts.prepare_batch_input_json_csv \
  --dataset_path ./datasets/PhysicalAI-Robotics-GR00T-GR1/gr1 \
  --metadata_path ./datasets/PhysicalAI-Robotics-GR00T-GR1/metadata.csv \
  --save_path ./sampling_dataset/PhysicalAI-Robotics-GR00T-GR1/gr1/pisa \
  --output_path ./sampling_dataset/PhysicalAI-Robotics-GR00T-GR1/gr1/pisa/batch_input.json


torchrun -m models.dreamgen.examples.video2world_gr00t \
  --model_size 14B \
  --gr00t_variant gr1 \
  --batch_input_json ./sampling_dataset/PhysicalAI-Robotics-GR00T-GR1/gr1/dense/batch_input.json \
  --disable_guardrail \
  --num_gpus ${NUM_GPUS} \
  --use_pisa \
  --pisa_density 0.2 \
  --pisa_block_size 64 \
  --pisa_start_layer_idx 2

### DreamDojo Action Conditioned
export PYTHONPATH="./models/DreamDojo"
export CUDA_VISIBLE_DEVICES=0 
export NUM_GPUS=1

# First, prepare the batch input JSON just like in dreamgen
python -m scripts.prepare_batch_input_json_csv \
  --dataset_path ./datasets/PhysicalAI-Robotics-GR00T-GR1/gr1 \
  --metadata_path ./datasets/PhysicalAI-Robotics-GR00T-GR1/metadata.csv \
  --save_path ./sampling_dataset/PhysicalAI-Robotics-GR00T-GR1/gr1/dreamdojo_dense \
  --output_path ./sampling_dataset/PhysicalAI-Robotics-GR00T-GR1/gr1/dreamdojo_dense/batch_input.json

# Run inference with batch_input_json
python -m models.DreamDojo.examples.action_conditioned \
  -o ./output/dreamdojo/cache_worldcache \
  --checkpoints-dir ./checkpoints/dreamdojo/2B_GR1_post-train \
  --experiment dreamdojo_2b_480_640_gr1 \
  --save-dir ./output/dreamdojo_cache/worldcache \
  --num-frames 49 \
  --num-samples 65 \
  --dataset-path ./datasets/PhysicalAI-Robotics-GR00T-GR1/gr1 \
  --data-split test \
  --deterministic-uniform-sampling \
  --batch-input-json ./sampling_dataset/PhysicalAI-Robotics-GR00T-GR1/gr1/dreamdojo_dense/batch_input.json




python -m models.DreamDojo.examples.action_conditioned \
  -o ./output/action_conditioned/cache_worldcache \
  --checkpoints-dir ./checkpoints/dreamdojo/2B_GR1_post-train \
  --experiment dreamdojo_2b_480_640_gr1 \
  --save-dir ./output/dreamdojo_cache/worldcache \
  --num-frames 49 \
  --num-samples 65 \
  --dataset-path ./datasets/PhysicalAI-Robotics-GR00T-Teleop-GR1/GR1_robot \
  --data-split test \
  --deterministic-uniform-sampling