This dataset contains 250 training cases for the real-world ablation of InstructVLA.
Beyond Collection Range
Written-out Numbers
Matchstick Numbers
Hard Case 1
Hard Case 2
All evaluation cases are excluded from the training set. For details, refer to the real_math.json for the training cases.
By co-training with a general multimodal dataset, we observe that InstructVLA performs better… See the full description on the dataset page:
https://huggingface.co/datasets/ShuaiYang03/Instructvla_realworld_math.