ShuaiYang03/instructvla_pretraining_v2_query_64_lora

Model Card for InstructVLA-SimplerEnv

Weights

• checkpoints/step-034500-epoch-08-loss=0.1167.pt: Complete model checkpoint for direct evaluation.
• checkpoints/step-034500-epoch-08_unload_lora.pt: Model containing only the VLM backbone and action expert. This cannot be used independently; it is intended solely for initializing Stage-2 training.
• checkpoints/pretrained_action_adapter: Action LoRA adapter. This cannot be used independently; it is intended solely for initializing Stage-2 training.

Evaluation Results

• results_step-034500-epoch-08-loss=0.1167_instruct_1-3: Results on SimplerEnv-Instruct.
• results_step-034500-epoch-08-loss=0.1167_1-3: Results on SimplerEnv.