an issue in running factscore_eval.py with mode = prompt-contrastive-decoding

[hehebamei]

When I used this mode, I got the bug like this:

• File "/data-store/shangguanzixuan/project/ICD-main/src/benchmark_evaluation/factscore_eval.py", line 225, in
  model_completion, c_dist = llm.generate(prompt, prompt_evil, **generate_kwargs)
  File "/data-store/shangguanzixuan/project/ICD-main/src/decoding_algorithm/contrastive_decoding.py", line 176, in generate
  outputs = self.model.generate(input_ids, evil_input_ids=evil_input_ids, max_length=max_len, num_return_sequences=1,
  File "/data-store/shangguanzixuan/anaconda3/envs/icd/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
  return func(*args, **kwargs)
  File "/data-store/shangguanzixuan/project/ICD-main/transformers/src/transformers/generation/utils.py", line 1756, in generate
  return self.prompt_contrastive_decoding_sample(
  File "/data-store/shangguanzixuan/project/ICD-main/transformers/src/transformers/generation/utils.py", line 3780, in prompt_contrastive_decoding_sample
  evil_outputs = self(
  File "/data-store/shangguanzixuan/anaconda3/envs/icd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
  return forward_call(*args, **kwargs)
  File "/data-store/shangguanzixuan/project/ICD-main/transformers/src/transformers/models/llama/modeling_llama.py", line 821, in forward
  outputs = self.model(
  File "/data-store/shangguanzixuan/anaconda3/envs/icd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
  return forward_call(*args, **kwargs)
  File "/data-store/shangguanzixuan/project/ICD-main/transformers/src/transformers/models/llama/modeling_llama.py", line 659, in forward
  position_ids = position_ids.view(-1, seq_length).long()
  RuntimeError: shape '[-1, 177]' is invalid for input of size 105

I guess you didn't reset the information in the Class LlamaForCausalLM (such as position_ids) when you passed in two prompts, which caused this error.