Changing the allowed maximum conversation length in Plato-2

[Sidx-sys]

Plato-2 removes all the conversation data points which exceed its constraint of conversation length. How can I increase this allowed conversation length so as to incorporate context lengths?

Please do help, I have been trying to do this for a while now. But everytime it seems like the pretrained weights that is being downloaded for the 24L model has a fixed unique vocabulary size, which leads to the training to not happen.

Thank you.

[sserdoubleh]

Using tool to expand position embedding: https://github.com/PaddlePaddle/Knover/blob/develop/knover/tools/expand_embedding.sh

sh knover/tools/expand_embedding ${PATH_TO_MODEL_DIR}

And then update training config file:
Add one line after

Knover/projects/PLATO-2/finetune/24L_train.conf

Line 30 in 6040f58

save_path="./output"

such as:

train_args="--max_src_len 384"