ricsinaruto/Seq2seqChatbots

Pre-trained model

shrenik007 opened this issue · 8 comments

Hi,
Can anyone tell me where can I find the pre-trained models or processed and checkpoints folder for this chatbot or help me by sending it?

After copying these trained models into the train_dir I just have to enter following command:

python t2t_csaky/main.py --mode=experiment

Right?

What will be in the decode_dir? I am getting following error while running python t2t_csaky/main.py --mode=decode command:

tensorflow.python.framework.errors_impl.NotFoundError: decode_dir/DailyDialog/trf_20_dropout-base/NCM_examples.txt; No such file or directory

it should be --mode=decode if you want to chat with the chatbot

What are the steps to setup these pre-trained models?

I just realized that the pre-trained models won't work with the current version of the repo, because they were trained with a much older version of tensor2tensor and tensorflow.
You can either try switching to the old version, but I honestly don't recommend this, or run your own trainings from scratch.

Alternatively I uploaded some newer trainings here.

The steps to setup the repo and run commands are detailed in the readme. Make sure to check the arguments in the config file before running anything.

Let me know if you have any specific questions.

The trained models you uploaded in the previous comment are Cornell, Twitter and DailyDialog. If I want to use Cornell then which problem I am suppose to use?

For Cornell you have to use the cornell_chatbot_basic problem