| ckpt_path=./exps/TTA/GRPO_flant5_44kMMVAE_fluxaudio_audiocaps_qwen25omni_semantic_numgpus2/model_1000.pth | |
| config_name=GRPO_flant5_44kMMVAE_fluxaudio_audiocaps_qwen25omni_semantic | |
| output_path=./output/GRPO_flant5_44kMMVAE_fluxaudio_audiocaps_qwen25omni_semantic_numgpus2 | |
| output_path="$(realpath -m "$output_path")" | |
| python eval.py \ | |
| --config ${config_name}.yaml \ | |
| --model_path "$ckpt_path" \ | |
| --output $output_path \ | |
| --num_steps 25 \ | |
| --cfg_strength 4.5 \ | |
| --eval_dataset tta-bench-acc | |
| cd ttabench | |
| bash run_eval.sh $output_path |