search

openai / gpt-2

Code for the paper "Language Models are Unsupervised Multitask Learners"
https://openai.com/blog/better-language-models/
Other
22.57k stars 5.53k forks source link

Generate text samples by giving some features or text as a parameter #217

Open kajlasolemn opened 4 years ago

kajlasolemn commented 4 years ago

I am using GPT2 small model (124M). I have trained the model on real Estate home descriptions samples. So it start generating homes description. I am use the GPT2 following generate function to generate home description samples. gpt2.generate(sess,length=500,temperature=0.9,nsamples=10,run_name='run1')

Well the generated description well look like this,

2700 NW Dogwood St Unit H104 is a condo in Seattle. WA 98146. This 830 square foot condo features 2 bedrooms and 2 bathrooms having 3 stories and 3 floors.This property was built in 1992 and last sold for $40000. .Based on Redfin's Seattle data. we estimate the home's value is $472830. Comparable nearby homes include 2700 NW Dogwood St Unit H101. 2700 NW Dogwood St Unit H202.

So my question is that how can i give the parameter like "4 stories 4 floors" to GPT2 generate function and it will generating the home description samples with this feature ("4 stories 4 floors") and if i give parameter like 2 bedrooms and 2 bathrooms it will start generating samples with this feature ( 3 bedrooms and 3 bathrooms )

giterdun345 commented 4 years ago

I have been trying to use the same sort of concept but for cover letters based on job descriptions. Have you made any headway on this yet?

drizzt00s commented 4 years ago

I have a question, can you make the generated description longer? such as character's length is 5000?

giterdun345 commented 4 years ago

I am almost certain you can make it as long as you want but keep in mind the longer the span of text that is generated the higher the likelihood of getting off topic or not remaining coherent. If you would want to do 5000 words I would iterate through this. Find a sweet spot with the longest span of words and generate, say around 250-350 words. Upon completing that iteration you can provide a new input to produce another 250-350 words and continue until...maybe you write a novel. You should test it out and see how many words you can generate until it falls off track. I am interested to see what you discover.

On Fri, Aug 21, 2020 at 11:23 PM drizzt00s notifications@github.com wrote:

I have a question, can you make the generated description longer? such as character's length is 5000?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/openai/gpt-2/issues/217#issuecomment-678592088, or unsubscribe https://github.com/notifications/unsubscribe-auth/ALCTCNABFN6ES6BBCLGYDXTSB5B2VANCNFSM4JZ7EPUQ .