DeepSpeedExamples
DeepSpeedExamples copied to clipboard
Assertion `srcIndex < srcSelectDimSize` failed
When I try to run Stage 3 finetuning PPO for qwen 2 0.5B model, I got the following bug:
Assertion srcIndex < srcSelectDimSize failed,
which seems like issue about input dataset sequence length?
I have already set Num_Padding_at_Beginning=0 # this is model related
@boqiny - do you have a full repro script? And can you share the device type and info about the system you're on?