It seems that video content is necessary? #237
Description
Hello, I am now running GRPO + lora in Qwen3_VL_8B_Instruct, but I met problems as follows:
Traceback (most recent call last):
File "/mnt/user3/TopicC/Qwen-VL-Series-Finetune/src/train/train_grpo.py", line 285, in <module>
train()
File "/mnt/user3/TopicC/Qwen-VL-Series-Finetune/src/train/train_grpo.py", line 259, in train
trainer.train()
File "/data/user3/miniconda3/envs/miqa/lib/python3.10/site-packages/transformers/trainer.py", line 2325, in train
return inner_training_loop(
File "/data/user3/miniconda3/envs/miqa/lib/python3.10/site-packages/transformers/trainer.py", line 2674, in _inner_training_loop
tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
File "/data/user3/miniconda3/envs/miqa/lib/python3.10/site-packages/transformers/trainer.py", line 4014, in training_step
inputs = self._prepare_inputs(inputs)
File "/data/user3/miniconda3/envs/miqa/lib/python3.10/site-packages/trl/extras/profiling.py", line 98, in wrapper
return func(self, *args, **kwargs)
File "/data/user3/miniconda3/envs/miqa/lib/python3.10/site-packages/trl/trainer/grpo_trainer.py", line 1067, in _prepare_inputs
generation_batch = self._generate_and_score_completions(generation_batch)
File "/mnt/user3/TopicC/Qwen-VL-Series-Finetune/src/trainer/grpo_trainer.py", line 160, in _generate_and_score_completions
prompt_inputs = self.processing_class(**processor_kwargs)
File "/data/user3/miniconda3/envs/miqa/lib/python3.10/site-packages/transformers/models/qwen3_vl/processing_qwen3_vl.py", line 170, in __call__
videos_inputs = self.video_processor(videos=videos, **output_kwargs["videos_kwargs"])
File "/data/user3/miniconda3/envs/miqa/lib/python3.10/site-packages/transformers/video_processing_utils.py", line 206, in __call__
return self.preprocess(videos, **kwargs)
File "/data/user3/miniconda3/envs/miqa/lib/python3.10/site-packages/transformers/video_processing_utils.py", line 372, in preprocess
videos, video_metadata = self._decode_and_sample_videos(
File "/data/user3/miniconda3/envs/miqa/lib/python3.10/site-packages/transformers/video_processing_utils.py", line 296, in _decode_and_sample_videos
videos = make_batched_videos(videos)
File "/data/user3/miniconda3/envs/miqa/lib/python3.10/site-packages/transformers/video_utils.py", line 216, in make_batched_videos
flat_videos_list = convert_pil_frames_to_video(flat_videos_list)
File "/data/user3/miniconda3/envs/miqa/lib/python3.10/site-packages/transformers/video_utils.py", line 165, in convert_pil_frames_to_video
if not (isinstance(videos[0], (list, tuple)) and is_valid_image(videos[0][0])):
IndexError: list index out of range
My training set is all image data without any video. However, the code seems to be trying to process video data, hence the error. How can I resolve this? Or does it currently only support mixed image and video input?