Beam Search Command Clarification

[tarsur909]

Hi,

Thanks for your great work!

Could you please clarify how to run {4, 16, 64, 256} compute budgets for beam search and DVTS to reproduce Figures 4, 5, and 7 in the paper. Where exactly does one specify the compute budget in each of the commands below?

Beam Search

cd src
bash scripts/run.sh --method beam_search --LM $POLICY_MODEL_PATH --RM $VALUE_MODEL_PATH --width 4 --num_seq 1

DVTS

cd src
bash scripts/run.sh --method beam_search --LM $POLICY_MODEL_PATH --RM $VALUE_MODEL_PATH --width 4 --num_seq 1 --num_q 64 --bs batch_size