Hi~ Thank you for your fascinating work!
May I please ask for the specific training configs when I sft the model with wod-e2e datasets (both with and without the CoT)? I am planning to reproduce the results in the Table S4 row 3 and row 4. Could you please give me some suggestions?
The training config I asked may include the learning rate (lr), the total batch size, the warm-up ratio, the weight lambda in the loss and other super parametres that differ from the configs in the [qwen2.5-vl-3B-mix-sft.yaml](https://github.com/ucla-mobility/AutoVLA/blob/main/config/training/qwen2.5-vl-3B-mix-sft.yaml).
Hi~ Thank you for your fascinating work!
May I please ask for the specific training configs when I sft the model with wod-e2e datasets (both with and without the CoT)? I am planning to reproduce the results in the Table S4 row 3 and row 4. Could you please give me some suggestions?
The training config I asked may include the learning rate (lr), the total batch size, the warm-up ratio, the weight lambda in the loss and other super parametres that differ from the configs in the
[qwen2.5-vl-3B-mix-sft.yaml](https://github.com/ucla-mobility/AutoVLA/blob/main/config/training/qwen2.5-vl-3B-mix-sft.yaml).