kevinng77 / blenderbot_paddle

用Paddle复现Recipes for building an open-domain chatbot论文

Hi this is kevin.

Domain: NLP (Large Language Model Training & Serving), Computer Vision (Conditional Image Gneration)

Interest: Front-End Design, Chatbot, AIoT

blenderbot_paddle's Issues

Please open a pull request to PaddleNLP to finish your submit.

Using other model for conversation

Hi, I tried to use this repo to do some conversation test, but when I change the model to small-90M, the results become weird. Here is the code I used for loading the model:

import paddle
from paddlenlp.transformers import BlenderbotSmallTokenizer,BlenderbotSmallForConditionalGeneration

model_name = "blenderbot_small-90M"

tokneizer = BlenderbotSmallTokenizer.from_pretrained(model_name)
model = BlenderbotSmallForConditionalGeneration.from_pretrained(model_name)

Then I use this code to enable the conversation:
print('Using model:' + model_name)

while True:
sample_text = input('User (input 'end' to end the conversation): \t')
if sample_text == 'end':
break

inputs = tokenizer(sample_text, return_attention_mask=True, return_token_type_ids=False)
inputs = {k: paddle.to_tensor([v]) for (k, v) in inputs.items()}

result_ids, scores = model.generate(input_ids=inputs['input_ids'],
max_length=60,
min_length=20,
decode_strategy='beam_search',
num_beams=10,
length_penalty=0.65)
for sequence_ids in result_ids.numpy().tolist():
print("bot:\t", tokenizer.convert_ids_to_string(sequence_ids))

Then I got these outputs:
Using model: blenderbot_small-90M
User (input 'end' to end the conversation): hello
bot: ET(T3%ip-ET(T3%ip-ET(T3%ip-ET(T3%ip-ET(T3�-ET(T3%ip-ET(T3%ip-ET(T3
User (input 'end' to end the conversation): end

Could you tell me what happened or what should I do ?

Thanks!