[Bug]: CLI --initial-caption only appears on first image, and twice

[SamTyurenkov]

What happened?

I git cloned master.

I used:

if GENERATE_CAPTIONS:
  print("\nGenerating captions")
  !source /content/OneTrainer/venv/bin/activate; which python; \
  /content/OneTrainer/venv/bin/python /content/OneTrainer/scripts/generate_captions.py \
  --model=BLIP2 \
  --initial-caption="$CAPTION_INIT, " \
  --caption-prefix="$CAPTION_PREFIX, " \
  --mode=replace \
  --sample-dir=$DATASET_FOLDER_IN_GDRIVE

I got:

1.txt - Masha, pretty 18yo girl wearing military soviet uniform, pretty 18yo girl wearing military soviet uniform, iphone wallpaper

2.txt - Masha, 1girl, solo, brown hair, pants, ass, long hair, looking at viewer, brown eyes, nature, looking back, outdoors, forest, jacket, day, lips, realistic, sunlight, uniform, tree, belt, smile, from behind, shirt, military, standing, breasts

What did you expect would happen?

1.txt - no duplicate initial-caption in the text and more of other features

2.txt - initial-caption to appear after the caption_prefix

Relevant log output

I used BLIP2

/content/OneTrainer/venv/bin/python
Loading checkpoint shards: 100% 2/2 [01:05<00:00, 32.76s/it]
100% 4/4 [00:02<00:00,  1.51it/s]

Generate and upload debug_report.log

report generation doesn't really work like that in colab, I'll do it later if needed, but I don't think it's related to my specific environment.