search

dreasysnail / POINTER

MIT License
112 stars 19 forks source link

How to generate multiple samples from a group of key words using "--type sampling" #16

Open etrigger opened 3 years ago

etrigger commented 3 years ago

For a set of given key words, how to create diverse text generation like the live demo?

dreasysnail commented 3 years ago

right. when doing the inference. You run the inference.py with an additional argument "--type sampling"

etrigger commented 3 years ago

@dreasysnail @guoyinwang In the demo (http://52.247.25.3:8900/), one can generate diverse text based on keywords as shown below. Right now, the argument '--type sampling' only generate one sentence. How can I create diverse text as shown below? Should I run the inference.py multiple times with different top_k, top_p, and temperature? Will the beam search method mentioned in the paper be released?

Generated from the live demo. Generated sentence 1 : he was then announced as the ceo of apple inc ltd . , the founder , and co lead , and co ceo respectively . , Generated sentence 2 : founded by former ceo and co founder of the founders , with the co founder , and current ceo , who currently lead a company . Generated sentence 3 : the company was founded by its ceo jeff smith , with former ceo , founder and co owner of facebook and served as the lead ceo . Generated sentence 4 : he is the former ceo , former president and the founder , former president and current lead the company . Generated sentence 5 : he is the co ceo and co founder of , co founder , and the former chairman and a lead executive of the company .