Can any PHP Markov libraries from github be used to make an article rewriter/paraphraser?

[Daemon]

I've never used Markov chains before, and I understand that they make spammy content. But, can they be used to rewrite/paraphrase existing content?

I installed the Pegasus Python library from HuggingFace, but my server only has a CPU, and with a CPU it can only do about 1,000 words per minute. I found that a GPU is about 60 times faster.

So Pegasus seems to require a lot of resources. I was hoping for something faster, that can work with a CPU, and is in PHP. I don't know enough about Markov chains to know if they might be suitable for this purpose.

Are there any Markov PHP libraries that might be suitable for this purpose? To rewrite/paraphrase existing content.

Or if anyone knows of any other open source option for rewriting in PHP or Python, that doesn't require as much resources as Pegasus and can work on a CPU, that would also be good to know about. Of course the quality would be poorer than Pegasus but that's fine.

 

[PinguSpy]

Do you mean 1000 words / min = 1 unique article?

Is that not powerful enough?

I can’t believe this kind of script exists, I wish I know python. :/

 

[Daemon]

Do you mean 1000 words / min = 1 unique article?

Is that not powerful enough?

I can’t believe this kind of script exists, I wish I know python. :/

Well it's not just that, it's that there is an overhead. It can take like 7-20 seconds to load the model. And I'd probably actually mostly use it for spinning short sentences. The overhead to load the model is really the bigger issue.

I was ideally looking for something of lower quality, but faster on a CPU without the same overhead to load a large model.

 

[Xyox]

Well it's not just that, it's that there is an overhead. It can take like 7-20 seconds to load the model. And I'd probably actually mostly use it for spinning short sentences. The overhead to load the model is really the bigger issue.

I was ideally looking for something of lower quality, but faster on a CPU without the same overhead to load a large model.

If you have some understanding of programming it's not that hard to edit the script so that it loads the model once and then let's you run it multiple times.

 

[Daemon]

If you have some understanding of programming it's not that hard to edit the script so that it loads the model once and then let's you run it multiple times.

Well yes that's true, but I don't intend to be running it at once or in batch. I would be using it irregularly.

 

[PinguSpy]

Well it's not just that, it's that there is an overhead. It can take like 7-20 seconds to load the model. And I'd probably actually mostly use it for spinning short sentences. The overhead to load the model is really the bigger issue.

I was ideally looking for something of lower quality, but faster on a CPU without the same overhead to load a large model.

Does the output nice compared to Quillbot?
I have similar issue with you. I got establish content.
The last thing I need right now is paraphrase. I definitely no need for "wrote from scratch" using GPT3

Why not just rent the GPU server then trough aws?

 

[Daemon]

Does the output nice compared to Quillbot?
I have similar issue with you. I got establish content.
The last thing I need right now is paraphrase. I definitely no need for "wrote from scratch" using GPT3

Why not just rent the GPU server then trough aws?

I'd prefer to not pay extra, plus I'd be using it programmatically so I'd have to come up with some programmatic way to activate AWS instances. And I think that would take even more overhead than just loading the model.

Quillbot seems to be better, though I've never used Pegasus before and I haven't tried messing around with the settings. A Pegasus expert could probably improve this. Here's somet test results:

Original text from wikipedia:
Mathematics is an area of knowledge that includes such topics as numbers (arithmetic and number theory), formulas and related structures (algebra), shapes and the spaces in which they are contained (geometry), and quantities and their changes (calculus and analysis).

Pegasus:
Themathematical subjects cover a variety of topics including numbers, arrhythmimental and a number theory, formulas and related structures (and many other subjects), equations, quantity, number systems, geometry, and various changes to spaces and their changes ($90)

Quillbot:
Numbers (arithmetic and number theory), formulas and related structures (algebra), shapes and the spaces in which they are contained (geometry), and quantities and their changes are all covered in mathematics (calculus and analysis).

Original text from wikipedia:
The earliest roots in the history of science can be traced to Ancient Egypt and Mesopotamia in around 3000 to 1200 BCE. Their contributions to mathematics, astronomy, and medicine entered and shaped Greek natural philosophy of classical antiquity, whereby formal attempts were made to provide explanations of events in the physical world based on natural causes.

Pegasus:
Ancient Egypt and Mesopotamia in 2000 to 1200 Age are thoughts to have shaped the natural antiquity in scientific thought and theory.

Quillbot:
The origins of science can be traced back to roughly 3000 to 1200 BCE in Ancient Egypt and Mesopotamia. Their contributions to mathematics, astronomy, and medicine influenced and impacted classical Greek natural philosophy, which sought to provide formal explanations for happenings in the physical world based on natural causes.

Original text from wikipedia:
Little is known about his life. Aristotle was born in the city of Stagira in Northern Greece. His father, Nicomachus, died when Aristotle was a child, and he was brought up by a guardian.

Pegasus:
Most of the background information found on how he came to be is tucked away in the archives of Stagira.

Quillbot:
His life is largely unknown. Aristotle was born in the northern Greek city of Stagira. Nicomachus, Aristotle's father, died when he was a youngster, and he was raised by a guardian.

 

[siolio]

Pegasus models are known to be memory-intensive and sizable models. best known for paraphrasing (sequence2sequence),
since you mentioned hugging face, there provide an api access where you have the pegasus models hosted in their optimised infra, and can interfere with it thru the api.
just read thru their docs https://huggingface.co/inference-api

 

[watchintv]

This is all very interesting. But if I did this and bought my own backlinks, would I see any kind of results in terms of ranking? Has anybody done this and ranked?