JI, Ziwei

I occurred an error when I run guess.py with GPU, but it's ok to run guess.py with CPU. ``` CUDA_VISIBLE_DEVICES=1 python guess.py --class_type gender --model_type inception \ --model_dir models/21936/ \...

Output of tokenizer.encode has None for space

1

I downloaded the pretrained and fine-tuned model from https://s3.amazonaws.com/models.huggingface.co/transfer-learning-chatbot/finetuned_chatbot_gpt.tar.gz tokenizer_class, model_class = GPT2Tokenizer, GPT2DoubleHeadsModel tokenizer = tokenizer_class.from_pretrained(args.model_checkpoint) tokenizer.encode('good morning') The output is [3454, None, 1054, 40164]. Get None for space...

For Smooth Inverse Frequency in Simple Sentence Similarity.ipynb In your code, merge sentences1 & sentences2 and remove_first_principal_component **together**. ``` embeddings.append(embedding1) embeddings.append(embedding2) embeddings = remove_first_principal_component(np.array(embeddings)) ``` However, in original code of...

According to _GPT2.py_, for Reinforcement Learning, you use a **BERTGen** model which is loaded from 'models/BERT_scorer_ep9.pt'. How did you train this model? Is it related to _parse_programs.py_ ? But in...

There is only generate_no_beam_search in [generate.py](https://github.com/nouhadziri/Neural-Path-Hunter/blob/main/src/dialkg/generate.py). Could you offer the code for beam search?

In your paper, you mentioned that" introduce the knowledge to the pretrained language model by **post-training** on knowledge-augmented data." In my opinion, **post-train** is different from finetune. According to paper...

Thanks for open-sourcing the dataset and code! It would be very beneficial for analysis to gain a step-by-step trajectory that includes the current_observation, action, and next_observation. I'm currently working with...