prakhar21/textaugmentation-gpt2

Fine-tuned pre-trained GPT2 for custom topic specific text generation. Such system can be used for Text Augmentation.

textaugmentation-gpt2's Issues

Multi label

Thank you for your sharing.I've been studying your code.
Now I want it use on multi-label,but I found your code and examples is on a single label.
If I want to use it on multi-label,how should I modify it?

Create sentences instead of words

I wanna predict sentences without calling generate function multiple times.

数据集

请问作者方便把数据集发给我吗。麻烦啦

Does this works with Chinese?

Does this works with Chinese text corpus? Thanks!

Facing issue in generate.py

---

RuntimeError Traceback (most recent call last)
in
93 TOKENIZER, MODEL = load_models(MODEL_NAME)
94
---> 95 generate(TOKENIZER, MODEL, SENTENCES, LABEL, DEVICE)

in generate(tokenizer, model, sentences, label, device)
47
48 next_token_id = choose_from_top_k_top_n(softmax_logits.to('cpu').numpy()) #top-k-top-n sampling
---> 49 cur_ids = torch.cat([cur_ids, torch.ones((1,1)).long().to(device) * next_token_id], dim = 1)
50
51 if next_token_id in tokenizer.encode('<|endoftext|>'):

RuntimeError: All input tensors must be on the same device. Received cpu and cuda:0

generate.py cur_ids = torch.cat([cur_ids, torch.ones((1,1)).long().to(device) * next_token_id], dim = 1) name 'device' is not defined

NameError: name 'device' is not defined ?How can I solve it, thank you

generate.py cur_ids = torch.cat([cur_ids, torch.ones((1,1)).long().to(device) * next_token_id], dim = 1)

I am facing issues in train.py and also in generate.py

Can you please reply back so that I can discuss the problems I am facing.

train.py --> model = model.to(device) --> Error: name model is not defined
generated.py --> for pre-trained model I am getting issues for mismatch and checkpoints

Regards

n is not been used

I am afraid that the value n in 'generate.py' line 44 and line 46 is not been used. Could you help? Thanks.

Removed

Apologies. Wrong post and I've closed the issue.