Open Muennighoff opened 1 year ago
We currently parallelize generations for tasks that require multiple generations for each problem like HumanEval and MBPP. So batch_size
should be increased when the number of candidate solutions n_samples
is higher than 1 which is not the case here.
👍 ; Would make sense to also support batch_size
to batch multiple examples even when n_sample
is 1, no?
Yes definitely! Especially since some benchmarks can have thousands of problems
I am interested to work on this! If nobody else is currently working on it :)
yes, feel free to work on it!
The below works when setting batch_size 1 🧐
Probably related: