8000
How to use the finetune model to infer? when i train a single task, after 5000 steps, it generate these files, how i use these files to infer? · Issue #10 · HongzheBi/H_RDT · GitHub
You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab
8000
or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
How to use the finetune model to infer? when i train a single task, after 5000 steps, it generate these files, how i use these files to infer? #10
See P66-P87(self.policy = HRDTRunner.from_pretrained(...) in "H-RDT/inference/robotwin2_example/H-RDT/deploy_policy.py", you need to create a folder like checkpoint-5k and put in the "config.json" and "pytorch_model.bin" from 5k checkpoints, which are needed to load weights during inference.
In principle, the HRDTRunner class of hrdt_runner.py inherits from hub_mixin.py(available in both H-RDT/utils and H-RDT/inference/robotwin2_example/H-RDT/utils folders), This file has the "from_pretrained" method to load Pytorch pretrained weights and return the loaded model.
Activity
HongzheBi commentedon Aug 28, 2025
See P66-P87(self.policy = HRDTRunner.from_pretrained(...) in "H-RDT/inference/robotwin2_example/H-RDT/deploy_policy.py", you need to create a folder like checkpoint-5k and put in the "config.json" and "pytorch_model.bin" from 5k checkpoints, which are needed to load weights during inference.
HongzheBi commentedon Aug 28, 2025
In principle, the HRDTRunner class of hrdt_runner.py inherits from hub_mixin.py(available in both H-RDT/utils and H-RDT/inference/robotwin2_example/H-RDT/utils folders), This file has the "from_pretrained" method to load Pytorch pretrained weights and return the loaded model.