[Bug] internvl_hf: IndexError when visuals is empty list (video-only inputs)

[jong980812]

Checklist

• I have searched for similar issues before opening this one.
• I am using the latest version of lmms-eval.

Bug Description

Bug Description

When running internvl_hf with video-only tasks (no images),
generate_until passes an empty list [] to self.processor(images=visuals, ...),
which causes an IndexError inside the processor.

Error Traceback

File "lmms_eval/models/chat/internvl_hf.py", line 270, in generate_until
    inputs = self.processor(images=visuals, ...)
...
File "transformers/image_transforms.py", line 917, in group_images_by_shape
    device = images[0][0].device if is_nested else images[0].device
IndexError: list index out of range

Root Cause

videos is already handled with a None guard, but visuals and image_sizes are not:

# Only videos is guarded — visuals is not
if len(videos) == 0:
    videos = None

# This also crashes when visuals is []
gen_kwargs["image_sizes"] = [visuals[idx].size for idx in range(len(visuals))]

Fix

# Before
if len(videos) == 0:
    videos = None

inputs = self.processor(
    images=visuals,
    ...
)

gen_kwargs["image_sizes"] = [visuals[idx].size for idx in range(len(visuals))]

# After
if len(videos) == 0:
    videos = None
if len(visuals) == 0:
    visuals = None

inputs = self.processor(
    images=visuals,
    ...
)

gen_kwargs["image_sizes"] = [visuals[idx].size for idx in range(len(visuals))] if visuals is not None else []

Environment

• lmms-eval version: main branch
• Model: internvl_hf with OpenGVLab/InternVL3-8B-hf
• Task type: video-only (no image inputs)

Steps to Reproduce

.

Error Message / Traceback

.

Environment

.

Additional Context

No response