Closed whuFSN closed 4 years ago
Hi there, I couldn't reproduce your error. Could you use "pytorch=1.4" and "transformers=2.6", and make sure you unzip all_csv.zip. From the output, it seems the model is not properly loaded. Could you make sure the model.load_state_dict is actually working?
Thank you for your answers. I update pytorch to 1.4 and transformers to 2.6. It solved my problem.
Hello For GPT-2.py model, I decode the sentences with the script
CUDA_VISIBLE_DEVICES=0 python GPT2.py --do_test --load_from models/GPT_ep8.pt
but I got the outputs GPT_gpt2_0.03.json which has a low bleu_3, the content in the GPT_gpt2_0.03.json like this:"2-18424778-6.html.csv": [ "", "", "", "", "" ], "2-1458666-4.html.csv": [ "", "", "", "wins iaa mwcwcwcwcwcwcwcwcwcwcw", "" ]
Does it because of args.model==gpt2? Or it should be gpt2-medium