Hugging Face's logo

doanh25032004
/
video_gen_physics_backup

Model card Files
xet
 Community
video_gen_physics_backup / docs /sampling_dataset_dreamdojo.md
doanh25032004's picture
doanh25032004
Backup source tree of video_gen_physics (2026-07-31T14:21:08Z)
ec0a9aa verified
preview code
|
Raw
History Blame Contribute Delete
3.36 kB

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