BUG: when I test mme on internvl2, an error will occur：'InternVL2' object has no attribute 'accelerator'

[daixiangzi]

CUDA_VISIBLE_DEVICES="7"  accelerate launch \
    --num_processes=1 \
    --main_process_port 12348 \
    -m lmms_eval \
    --model internvl2 \
    --model_args pretrained="OpenGVLab/InternVL2-8B,device_map=auto" \
    --tasks mme \
    --batch_size 1 \
    --log_samples \
    --log_samples_suffix llava_mme \
    --output_path ./logs/

lmms-eval/lmms_eval/models/internvl2.py

Line 180 in bd9908f

elif accelerator.num_processes == 1 and device_map == "auto":

lm model have't self.accelerator = accelerator, but lm.accelerator.num_processes attribute will appear in evaluator.py,

lmms-eval/lmms_eval/evaluator.py

Line 639 in bd9908f

while len([file for file in os.listdir(cli_args.output_path) if file.endswith('metric_eval_done.txt')]) < lm.accelerator.num_processes:

[kcz358]

Oh, I think we forgot to add the self.accelerator when num_process is equal to 1. I will fix it later

[skyil7]

I think we can close this issue since @kcz358 fixed it.

1703b02

[daixiangzi]

I think we can close this issue since @kcz358 fixed it.

1703b02

yes