ucsd-ai4h/medical-dialogue-system

medical-dialogue-system's Issues

data link

hello, I can not open the link of train_data.json. Could you please update it?

Data is not avaliable

Hello, read access to the data on the link https://drive.google.com/drive/folders/1g29ssimdZ6JzTST6Y8g6h-ogUNReBtJD?usp=sharing is forbidden. Can you please update the link?

download link has been missing

download link has been missing

大佬，能不能把爬haodf.com的代码分享一下，感谢

大佬，能不能把爬haodf.com的代码分享一下，感谢

请问gpt2的预训练模型在哪里可以获取

The License of the Dataset.

Thanks for the extraordinary work!

May I ask the License for the dataset please?

数据来源

你好我想请问医患的数据是从网络上爬取的，还是真实的医院中的医患对话场景？因为我看到多轮中存在一些答非所问和错误的情况，期望得到您的回复，谢谢。

Requests for additional processed data (i.e., MedDialog-EN)

Hi,

Can I share the additional processed data of MedDialog-EN?

Thanks

some details

Hello,
it seems there are some problems?
in bertGPT/generate.py, the past_length should start from 0 following the training settings.

length = 1
# decoding loop
for i in range(100):
    mask = F.pad(mask, (0, 1), "constant", 1.0)
    logits, past = decoder(prev_pred, mask, past=past, past_length=length)
    logits = logits.squeeze(1)
    logits = top_k_logits(logits, k=top_k)
    probs = F.softmax(logits, dim=-1)
    prev_pred = torch.multinomial(probs, num_samples=1)
    sentence.append(prev_pred)
    if prev_pred[0][0] == 102:
        break
    length += 1

in gpt2/gpt2_test_mmi.py, the sequence curr_input_tensors should be truncated from the front.

curr_input_tensors = curr_input_tensors[:, 0:300]   ×
curr_input_tensors = curr_input_tensors[:, -300:]   √

When running the gpt2_train process,lack of tokenized file

So there to download train_med_tokenized.txt?