A dimension problem was encountered while running prep_sgni.py:

Hi @thu-coai @hzhwcmhf @MaLiN2223 @zqwerty @xiaotianzi @truthless11 and thanks so much for making your code available.
I want to fine tune the code on a new dataset that the format is very similar to IMDB dataset (it has a couple of sentences and label is positive/negative/neutral). Could you please advise on what changes I need to make?

I appreciate your time and help :).

您好， 看到您的modeling_roberta.py文件对modeling_bert有所依赖（例如：第175行）， 可否麻烦您提供一下modeling_bert.py文件？

I have read your paper and found it the label-aware masked-language model pretraining with SentiWordNet integration very interesting. I am curious to see how it is implemented practically for my research.

I was wondering if you had any plans to publish the code of this part of the research?

ValueError: Unrecognized model in ../SentiLARE. Should have a model_type key in its config.json, or contain one of the following strings in its name: fnet, gptj, layoutlmv2, beit, rembert, visual_bert, canine, roformer, clip, bigbird_pegasus, deit, luke, detr, gpt_neo, big_bird, speech_to_text_2, speech_to_text, vit, wav2vec2, m2m_100, convbert, led, blenderbot-small, retribert, ibert, mt5, t5, mobilebert, distilbert, albert, bert-generation, camembert, xlm-roberta, pegasus, marian, mbart, megatron-bert, mpnet, bart, blenderbot, reformer, longformer, roberta, deberta-v2, deberta, flaubert, fsmt, squeezebert, hubert, bert, openai-gpt, gpt2, transfo-xl, xlnet, xlm-prophetnet, prophetnet, xlm, ctrl, electra, speech-encoder-decoder, encoder-decoder, funnel, lxmert, dpr, layoutlm, rag, tapas, splinter

Because I want to use it to get sentence embedding

Hi,

I have downloaded the pretrained model and the preprocessed data(*_newpos.txt) from google drive.
For SST the reported val result is 55.04% in Table 15 and i get the performance 56.04%. however i only can get 84.92% on IMDB but the reported performance if 95.96% in Table 15.
And when i finetune the released model for downstream emotion recognition tasks, the results is poor than Bert-based model and RoBerta-Base model.

Is there any suggestions of these problems? Thanks!

for lr in 2e-5 do data_dir=/data7/emobert/resources/pretrained/sentilare/raw_data/sent/sst/ CUDA_VISIBLE_DEVICES=${gpuid} python run_sent_sentilr_roberta.py \ --cvNo 1 \ --data_dir ${data_dir} \ --model_type roberta \ --model_name_or_path ${pretrain_model_dir} \ --task_name sst \ --do_train \ --do_eval \ --max_seq_length 50 \ --per_gpu_train_batch_size 32 \ --per_gpu_eval_batch_size 32 \ --learning_rate ${lr} \ --num_train_epochs 8 \ --warmup_steps 100 \ --eval_all_checkpoints \ --overwrite_output_dir \ --seed 42 \ --output_dir ${output_dir}/${corpus_name}_roberta_base_finetune_lr${lr}_bs32_m2 done