MEMORY_TOKEN used but no memory_features #80
Description
Description
Hi, thanks for releasing StreamVLN.When running evaluation on my machine,something wrong happened:
Traceback (most recent call last): File "/mnt/data2/stream_ws/StreamVLN/streamvln/streamvln_eval.py", line 588, in <module> eval() File "/mnt/data2/stream_ws/StreamVLN/streamvln/streamvln_eval.py", line 538, in eval evaluate(model, tokenizer, args) File "/mnt/data2/stream_ws/StreamVLN/streamvln/streamvln_eval.py", line 557, in evaluate sucs, spls, oss, ones, ep_num = evaluator.eval_action(get_rank()) File "/mnt/data2/stream_ws/StreamVLN/streamvln/streamvln_eval.py", line 333, in eval_action outputs = self.model.generate(**input_dict, do_sample=False, num_beams=1, max_new_tokens=10000, use_cache=True, return_dict_in_generate=True, past_key_values=past_key_values) File "/home/aloha/miniconda3/envs/streamvln/lib/python3.9/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context return func(*args, **kwargs) File "/mnt/data2/stream_ws/StreamVLN/streamvln/model/stream_video_vln.py", line 398, in generate ) = self.prepare_inputs_labels_for_multimodal( File "/mnt/data2/stream_ws/StreamVLN/streamvln/model/stream_video_vln.py", line 232, in prepare_inputs_labels_for_multimodal raise RuntimeError( RuntimeError: MEMORY_TOKEN used but no memory_features for batch 0
This happened when I ran the evaluation on RXR,but before that ,the evaluation on the R2R was successful,whis has confused me for a long time.