Apologize if I misread the paper. When training Tk-instruct, did you use max_num_instances_per_task=100 or max_num_instances_per_task=64 to report the main results in Table 3 and Figure 4? I cannot find anywhere in the paper how many training instances per task are used. Figure 5(b) mentions that 64 instances per task give the best results, but in the code, the default setting is 100, which confuses me.
Hi, thank you for your great work!
Apologize if I misread the paper. When training Tk-instruct, did you use
max_num_instances_per_task=100
ormax_num_instances_per_task=64
to report the main results in Table 3 and Figure 4? I cannot find anywhere in the paper how many training instances per task are used. Figure 5(b) mentions that 64 instances per task give the best results, but in the code, the default setting is 100, which confuses me.Thank you!