Evaluation Metric For MI Dataset

[lsyysl9711]

The infer.py works smoothly for Esconv dataset while it is stuck in another MI dataset.

So in this line of code, we need to use one slot called "situation". But actually, there is no slot called "situation":

[lsyysl9711]

CUDA_VISIBLE_DEVICES=3 python infer.py \ --config_name strat \ --inputter_name strat \ --data_name mi \ --knowledge_name sbert \ --add_nlg_eval \ --add_mi_analysis \ --seed 13 \ --load_checkpoint "/mnt/storage/lsy/gdown/blenderbot/DATA/strat.strat.mi.sbert/2023-09-22111813.3e-05.16.1gpu/epoch-2.bin" \ --fp16 false \ --max_input_length 256 \ --max_decoder_input_length 40 \ --max_length 40 \ --min_length 15 \ --infer_batch_size 2 \ --infer_input_file ./_reformat/ \ --temperature 0.7 \ --top_k 30 \ --top_p 0.3 \ --num_beams 1 \ --repetition_penalty 1 \ --no_repeat_ngram_size 3

Here is my inference script.

[lsyysl9711]

I might find out the possible reason behind this: in MI dataset, we do not have a slot called "situation". So how should we fix this problem?