You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In verl/workers/reward_model/megatron/reward_model.py
# split into micro-batches
if self.config is not None and 'ppo_micro_batch_size_per_gpu' in self.config:
infer_batch_size = self.config.ppo_micro_batch_size_per_gpu
else:
infer_batch_size = data.batch.batch_size[0]
It seems use ppo_micro_batch_size_per_gpu instead of micro_batch_size_per_gpu, so reward_model.micro_batch_size_per_gpu not work, always infer_batch_size = data.batch.batch_size[0].
The text was updated successfully, but these errors were encountered:
In
verl/workers/reward_model/megatron/reward_model.py
It seems use
ppo_micro_batch_size_per_gpu
instead ofmicro_batch_size_per_gpu
, soreward_model.micro_batch_size_per_gpu
not work, alwaysinfer_batch_size = data.batch.batch_size[0]
.The text was updated successfully, but these errors were encountered: