MMaDA/script/eval_t2s_grid.sh

#!/bin/bash

timesteps=(12)
speech_token_lengths=(150)
train_steps=($(seq 135000 -50000 35000))

for ts in "${timesteps[@]}"
do
  for stl in "${speech_token_lengths[@]}"
  do
    for step in "${train_steps[@]}"
    do
      echo "=========================================================="
      echo "Running with: train_step=${step}, timesteps=${ts}, speech_token_length=${stl}, guidance_scale=0.75"
      echo "=========================================================="

      torchrun --nproc_per_node=8 MMaDA/inference_t2s_emova.py \
              config=MMaDA/configs/mmada_demo_s2t.yaml \
              --train_step ${step} \
              --guidance_scale 0.75 \
              --timesteps ${ts} \
              --speech_token_length ${stl}
    done
  done
done

echo "Grid search finished."