Closed ajayvohra2005 closed 9 months ago
Thank you for reporting the issue. Apologies for the confusion with the variable. We will change the variable name to NUM_NEURONCORES=32. However, one thing to note, that variable is used only in that shell script and can be named to anything.
Samples shell scripts' variables confuse devices and cores. A
Trn1
instances has 16 Neuron Devices (chips), each with 2 cores.This sample script, on Line 31 shows:
export NEURON_NUM_DEVICES=32
I think, the correct code would be:
export NEURON_NUM_CORES=32
The deprecated Neuron Megatron example script shows it correctly:
NUM_NEURONCORES=32