Skip to content

Latest commit

 

History

History
26 lines (17 loc) · 1.35 KB

File metadata and controls

26 lines (17 loc) · 1.35 KB

Dynamics Fintuning

Processing Real-World Data

First, aggreage real wold data with the following, replacing <dataset_name> with the name of the dataset in datasets/real:

python scripts/aggregate_realworld_data.py dataset_name=<dataset_name>

Next, process the data the same way as in Data Generation, with:

python scripts/process_data.py dataset_name=<dataset_name>

Finetuning the World Model

Finetune the dynamics only of the world model with the below, replacing <dataset_name> with the name of the dataset in datasets/real/ and <resume_checkpoint> with the name of the checkpoint to finetune from checkpoints/models/. Checkpoints will be saved in checkpoints/models/; use <run_name> to specify the name of the checkpoints (optional; it's recommended to use a different name than the original checkpoint). Optionally, launch with wandb.log=true and wandb.entity=<your_wandb_entity_or_username> to enable Weights and Biases logging. Additional configuration is found in config/finetune_model.yaml.

python scripts/finetune_model.py data.dataset_name=<real_dataset> checkpoint.resume_checkpoint=<resume_checkpoint> run_name=<run_name>