openai gpt-2

21/03/2019

Another openai project I’ve been playing with is gpt-2:
Code for the paper “Language Models are Unsupervised Multitask Learners

Synopsis, you give the bot a text extract and it responds 🙂
I’ve been through the process of creating a local Docker and running this script, I’m far from understanding the complexities bit here is a demo:
python3 src/interactive_conditional_samples.py --top_k 40

Model prompt >>> I’ve been through the process of creating a local Docker and running this script, I’m far from understanding the complexities bit here is a demo:

======================================== SAMPLE 1 ========================================
http://davidh.com/docker-install-file.jpg

When Docker is launched, you will see a log of the server as part of its initial state, and the directory that this Dockerfile is created on:

$ curl -L https://davidh.com/docker-install-file.jpg && sudo sh –dump –remote | tail -n3

(This might surprise you, I know, but I’ll keep this simple)

If you are a system with Docker image installed that is running a version of Linux you can easily see that these will do the job.

You can also check the process from that point:

$ curl -L https://davidh.com/docker-install-file.jpg && sudo sh –dump –remote && sudo echo ‘running in directory: %s’ % { ‘username’: ‘[email protected]’, ‘password’: ‘[email protected]’ }

You can view this process in the documentation: http://davidh.com/guide.html

To run the daemon, you have to: http://davidh.com/guest.html

To run all those other tasks (it’s all within the shell)

$ sudo bash

To run a shell script to run the daemon, there are two options:

$ sudo bash $ sudo crontab

To run a shell script which takes some of the shell commands (such as ‘run-sh’) you know.

$ sudo bash –help

Here’s the full description:

The bash command line option specifies that the shell shell function is expected to run when run. It specifies an option that specifies how to run the shell, and what that will do. See help/shell for details.

The following variables are included:

pst_param – The current local path of the process whose pst_param parameter is being used. To see the current state and current PID, see PID.

– The current local path of the process whose parameter is being used. To see the current state and current PID, see PID . tty – The system variable you will use to set the tty.

– The system variable you will use to set the tty. commandfile – Specifies the line-of-the-art file to run. This file may be a non-

mmm let’s try a less technical input, maybe an extract from a newspaper article:
Christchurch attacks: New Zealand brings in sweeping gun-law changes

Model prompt >>> “I absolutely believe there will be a common view amongst New Zealanders, those who use guns for legitimate purposes, and those who have never touched one, that the time for the mass and easy availability of these weapons must end. And today they will,” said Ardern.

======================================== SAMPLE 1 ========================================

‡There will be an eventuality, if you put into practice any basic principle, which I have spoken about before: a lack of awareness on the part of the security agencies, and the lack of concern by both community stakeholders and law enforcement, who believe this has to do with the fact that this is a firearm and a person and an environment that should take priority over law enforcement, ‒ which, of course, this has not done.

‡Let’s take this for example, for example when Mr Abbott says he is going a ‘war on gun violence’, or when some on the right have been saying that there are millions of people murdered each year and millions of kids under 12, and that the government should consider making those numbers into figures like that.

‡But I think if the government thought this was a real public health issue, and was interested in making them a reality, those are very different things from people who are interested in a universal public health problem. Why should we want to impose that on the people who are killing people and the kids, especially those who have no means to move forward in their lives, no safety net, no other options? It looks like the government has been making more changes to the guns, so I’m not sure what the government is now looking to do.

‡What I do think is that when people make that argument. When people say that the people that have been suffering the most in their life, and I think that was a common denominator between those two statements, and that the violence would be caused by someone carrying and killing themselves, and that the violence would never get resolved to an end on paper, then I think that’s an incorrect position, a misunderstanding of what’s going on. And it’s not something that’s going to get solved, right?


‡Mr Abbott is wrong on that, and that’s for Mr Ardern to get into.

‡I would like the Government to continue to do that.

‡It seems a lot easier to address our mental health problems where our health care would have been better if the legislation was in place. For a lot of the people who are doing that, mental health is not the primary concern, or the primary concern of these issues. It’s not that they want to live in a mental health regime for their families. It’s that they want to be treated for a specific issue.‡ When the government is really interested in fixing

Interesting and curious eh?
I’ll continue to dig into this paper, see what else I can comprehend, and fingers crossed one day we might see the full release of this project shared 🙂

Thanks heaps to those who have worked and shared this!

citation

@article{radford2019language,
title={Language Models are Unsupervised Multitask Learners},
author={Radford, Alec and Wu, Jeff and Child, Rewon and Luan, David and Amodei, Dario and Sutskever, Ilya},
year={2019}
}