question on custom DROID dataset LoRA finetuning on pi0.5-driod

[kim-hansol314]

I was looking for a config to train on custom DROID dataset LoRA finetuning on pi0.5-driod
but it seems that it is not provided.

    TrainConfig(
    name="pi05_droid_lora_finetune",
        model=pi0_config.Pi0Config(
            pi05=True,
            action_dim=32,
            action_horizon=16,
            paligemma_variant="gemma_2b_lora",
            action_expert_variant="gemma_300m_lora",
        ),
        data=LeRobotDROIDDataConfig(
            repo_id="hskim314/my_droid_dataset",
            base_config=DataConfig(prompt_from_task=True),
            assets=AssetsConfig(
                assets_dir="gs://openpi-assets/checkpoints/pi05_droid/assets",
                asset_id="droid",
            ),
        ),
        weight_loader=weight_loaders.CheckpointWeightLoader("gs://openpi-assets/checkpoints/pi05_droid/params"),
        num_train_steps=20_000,
        batch_size=32,
        freeze_filter=pi0_config.Pi0Config(
            pi05=True,
            action_dim=32,
            action_horizon=16,
            paligemma_variant="gemma_2b_lora",
            action_expert_variant="gemma_300m_lora",
        ).get_freeze_filter(),
        ema_decay=None,
    )

i modified pre-defined config to train on it, but the result is not very satisfying.
is there anybody who can share the config that worked successfully?