Skip to content

Latest commit

 

History

History
49 lines (43 loc) · 1.18 KB

README.md

File metadata and controls

49 lines (43 loc) · 1.18 KB

MetaWorld_Environment

For training data and checkpoints, please refer to Data & Model Checkpoints. Please fill in OpenAI key in evaluate.py for online gpt access.

LanguageType

LanguageType Meaning
no No Language
h Hindsight
f Foresight
hf Hindsight & Foresight
rh GPT-augmented Hindsight
rf GPT-augmented Foresight
rhf GPT-augmented Hindsight & Foresight

Training

Train models for Hypothesis 1

source env.sh
source scripts/train/h1/train_{LanguageType}.sh

Train Models for Hypothesis 2

Pretraining

source env.sh
source scripts/train/h2/pretrain/pretrain_{LanguageType}.sh

Adaptation

Please modify the load_ckpt path and the shot_number in the script.

source env.sh
source scripts/train/h2/adapt.sh

Evaluation

Evaluation for Hypothesis 1

source env.sh
source scripts/eval/eval_h1.sh

Evaluation for Hypothesis 2

source env.sh
source scripts/eval/eval_h2.sh