mllm_eval/ directory referenced in README is missing from the repository

[KetoneOlefine]

The evaluation/README.md and REPRODUCE_BAGEL_RESULTS.md both document an AC (Appearance Consistency) scoring pipeline that depends on a mllm_eval/ directory:

4. Run MLLM-based AC scoring (optional but recommended)
   Launch the evaluation helper in mllm_eval/ to produce AC results with a serving LLM:

cd mllm_eval
bash eval_infer.sh <model_path> default <port>

However, the mllm_eval/ directory (and its scripts build_lmdb.py, mllm_eval.py, eval_infer.sh) is not present in the released repository.

The only AC-related code available is eval/aggregate.py, which can read a pre-existing AC predictions JSON, but cannot produce one.

[Z-MU-Z]

Thanks for reporting this — you’re absolutely right for the previous release package.
We have now added the missing evaluation/mllm_eval/ directory (including build_lmdb.py, mllm_eval.py, and eval_infer.sh) to main.

[KetoneOlefine]

Thank you for the quick fix! I appreciate it.
I will sync with the latest main branch and test the AC evaluation pipeline.