as close as I've gotten with google colab

global tempuratures

Dout of chiss my child

cut than piss of rifs twiend worters a

wotching to see tham

lack softermo thent is

lopt awogc the mount in

as greepending the winder

canders and my thiser

on a staing autunf rain

muncainy ocing to mour

as

I coul sadory

moini...

Sadenive the grass

wat inses got hish-alas...

A thes fulling and blay sand...

Or is the light rove caser

A shanter grovering

at waters and street'd and

whire wetper-jass farmer in and

talk young bringer in the streat

Deriots mith-flaved of sand...

Twilight whippoorwill ...

Whistle on, sweet deepener

Reciting scriptures...

While I turned more

in houtt in to shimany

.... warmend so

A stingle butterfly

ford storching the samch and ...

A chilent in the witer

fropsed the a longer your not...

Coreced comes tomp pamuted

af incerels tadeniland

can alx outunn willowe!

Green stollen wind rins...

The scert-back scrarecring

farelow that is set

Dey shaping and dien

of this two snail budlead

fall ... The scery

In the fatt ofen

flower botce-ten the scerp's

whise betured by hot ...

But sticky in the air ...

Twing and swreen sturmer

So chilen

switelf crorching ...

Pretty butterfly

pearfus fref ingers an ... And fust

age

For morning-glories

I can foresee grave danger...

Can't it get away

from the sticky pine-branches ...

cilenes on the fores

bond bnim preants for love ...

Falls of it shiny wings and look...

Tiny sentences

brushing soft on my shutters...

Bush-dusky and crocked

A shitter now silent

The following text is as close as I've come to getting the output I want from text data in google colab.

Well I had to seperate some text about climate change into frases that I felt where poetic and then I trained on this data with CharRNN a well known generative text model (character level LSTM) created by Andrej Karpathy

It seems when I step away from the computer too long the website disconnects and I wind up losing data. I did 2 cycles of training but could have done more to get more coherent phrases. I did like one phrase

"a shitter no silent"

I may be able to press on and get something here... the idea was to train on poetry...get coherent phrase then switch the trained model to a different data set. What I had to do was take poetyr and some text about climate change and then mix them together myself...and throw them back in the model to train... This will work but hasn't yet...