it's nice to think about this algorithm part
i'm indoors right now and where i was working on it was outdoors. i'll
just open up an unsync'd copy here
that system is offline; i wonder how in the world i got my code onto
it O_O maybe i copied it over by hand? maybe i started dev on it? i
dont'
it seems it would be good to get something that functions, and uses
flush regions, just like how i started the old approach. even if it
doesn't have most of the things planned to include.
so very basic code would simply be to add writes as edges between flush regions
and not do any rebalancing at all. simply use the chunk region class
to subdivide them.
then we would move straight to the core inhibited part of retaining
binary-like trees that are distant from write areas.
so a write has two ends, a start and an end, and each one can either
be within an old write, within an old flush, or in a new area.
then similarly, they can span all or parts of writes and flushes, as well.
the old code uses start_idx and end_idx and then excises that range.
the ends are treated
it also makes sense to put them in one list, so one can quickly find
what is overlapped
the linked lists swap back and forth with regard to the types of data linked
it makes sense to have two different classes for these, so they can
have separate methods.
i wonder if i can use the adapter finetuning stuff to feel like i am
moving forward on the semibalanced trees a little
i'm imagining a model that mutates code in a way that is helpful:
stabilising it, simplifying it, anything helpful for the goal
if the code were small enough, it might fit right
long story short, from https://arxiv.org/pdf/2112.07916.pdf ,
transient global attention is a new approach to attention invented for
longt5, which appears to reliably outperform local attention in the
same architecture
the appearance of that, combined with finetunings i saw on the
huggingface hub
each flush region has a left write and a right write. these are passed
to the constructor.
a flush can happen two ways:
1 - new flush. starts with nothing in it. adds only writes.
2 - old flush. starts with a previous flush and no writes. new writes
can break up the reference to the previous
I'm a making a Write class to balance the concept of writes with flushes.
This helps make concepts around the design have less complexity for me.
maybe Chunk can be the linked list node, since everything that is a
chunk is a member of a linked list.
noting also that the list index is available when accessing the nodes,
so the link members might be mostly for simplification, or a nonverbal
possible plan i'm not thinking of much yet
here's a recent stable code. this was before the 'fiction' approach i
believe; classes could still be hoisted to be subclasses.
https://lists.cpunks.org/pipermail/cypherpunks/attachments/20220716/dec3d2ee/attachment-0001.py
https://www.youtube.com/watch?v=Ci2npsJIvFc Eben Alexander
DOPS at Virginia University.
Aware2 Study by Dr Sam Parnia at Southampton university.
Dr Bruce Greyson
Dr Raymond Moody Jr
Sir Roger Penrose
Stuart Hameroff
Paul Stamets
Antonio R. Damasio
Donald Hoffman
Conscious by Rita Carter
von
https://twitter.com/StandForBetter/status/1543290993046196224
it turns out adapter-transformers actually has a mutated entire copy
of the transformers repository inside it, and replaces this on the
user's system. they update this regularly, but their current version
does not have longt5.
i added longt5 and forked the repo and pushed it. it doesn't work with
so things are split between chunk regions and flush regions
makes for a stronger cognitive challenge, to change the source in
multiple concepts at once
i guess it's just inspiring to add in names that seem a little more intuitive.
it is of course confusing that both are 'regions', when juggling
i think where i left off, i was trying to kind of rewrite the add()
function to just plain use the new flush regions. without regard to
all the old techniques. simplification.
next step i think: find recent stable code, such as whatever was last posted
regarding thinking/spamming, could use more stabilisation around idea
of separating writes from flush regions into peer classes possibly
with different methods. this plan could be tuned more by planning the
i seem to be a little interested now in discerning if transformer xl
will run on my local system
there's a doc page at
https://huggingface.co/docs/transformers/model_doc/transfo-xl . it
starts off with this example:
from transformers import TransfoXLConfig, TransfoXLModel
# Initializing a
i had to install the transformers package from git in order to use LongT5Model
thats' likely something like pip3 install --upgrade
git+https://github.com/huggingface/transformers
the example code doesn't work, but the model loads when i put it in a
normal huggingface 'pipeline' class
it appears to do forward inference on my cpu, haven't figured out how
to put it on the gpu
there are many different kinds of complexity
i'm noting how the kinds i am handling are somewhat indicated by the
parts of the code i have improved vs left
having the two separate classes seems it would be helpful for design.
putting methods on them.
it looks like this is the smallest of the most official longt5 models:
https://huggingface.co/google/long-t5-tglobal-base
something i could theoretically try could be to manually process the
weights to reduce their precision
i'm a little confused around the difference in code paths between new
flushes and old flushes.
but, noting that adding writes to an empty flush is the same as adding
writes that are sparsely distant from the bounds of a previous flush.
i'm guessing the flush data will be a list of flush regions.
it seems the core behavior is adding a write. this branches various
conditions depending on whether it intersects a previous flush, a
previous write, opens rebalancing, etc.
the current written bounds are earliest where old_end >= new_start,
and earliest where old_start > new_end
earliest where old_end >= new_start : if this isn't off the right end,
then it intersects or touches the new data. if end == start, it
touches. if end > start, it overlaps, unless it is off
what happens if a write does not overlap a region ...?
maybe it would make sense to think of the region of note as being the
one prior to the upper index
i have so much semibalancing inhibition, that the linked lists get
inhibited too. i am already weak around linked lists due to graph
inhibition i experience.
so i'm mostly working on the linked lists now, and am presently
planning to balance the trees after i have a workable framework to do
so
so there are two factoring concepts here
one is the linked lists, for organising of functionality
the other is the semibalanced trees.
it seems it would make sense to give the flush regions and the writes
a common base class of being a node in a linked list.
i think there are also some parts that could optionally be removed or
simplified, since writes are a special thing used only in the latest
flush, they don't need to have
a nice thing here is that i have existing stable code
so as i conceptualise and lose things more, a workable approach may
trend toward changes to stable code, that keep the code stable.
i ended up searching huggingface's github for closed issues that
mentioned 'transformer-xl'
it showed me results for other long context transformers than just that one.
the most recent it showed completed i think was longt5.
i didn't find transformer-xl models in the huggingface hub; it was
not enough gpu ram
cool to learn about though
class Region:
def __init__(self, *entries, left_write = None, right_write = None):
self.entries = entries
self.left_write = left_write
self.right_write = right_write
class Write(Chunk):
def __init__(self, offset, data, flush, left_region
regarding thinking/spamming, could use more stabilisation around idea
of separating writes from flush regions into peer classes possibly
with different methods. this plan could be tuned more by planning the
semibalancing.
i am risking my focus in order to spend a little time learning what
the difference might be between local attention and transient global
attention in longt5
as a side note, it is notably that it looks like a little like people
are using longt5 heavily.
the model is 1G large
i'm suspecting this implies it won't fit in my gpu ram here
there appears to be useful information on expected ram requirements at
https://huggingface.co/docs/transformers/v4.16.2/en/performance#model-weights
long-t5-tglobal-base is using 1539 MB nvidia GPU ram for me with float32
weights. I unloaded my desktop environment and it would perform simple forward
inference for me.
He won the 2021 Superbike World Championship, ending Jonathan Rea's 6-year
long reign. Toprak is the first Turkish World Champion in the Superbike history.
https://en.wikipedia.org/wiki/Toprak_Razgatlıoğlu
So another night just sitting here being entrapmenped. The technology is
cool but at this stage it seems so petty.
xNY.io - Bank.org probably are awsome, that's why they would do this?
The FBI Cyberstalking team at the New York Field office have ignored me
when reported. So, that must mean sonic
Faculty Publications
UC Hastings College of the Law Library
Author: Roger C. Park
Title: The Entrapment Controversy
Source: Minnesota Law Review
Citation: 60 Minn. L. Rev. 163 (1976).
Brody Larson has made 212 highlights:
Brody Larson has been hearing that they maliciously conspired to weave a
false narrative that his dad was, was colluding with a hostile foreign
sovereignty (North Cyprus).”
Brody has alleged
the readme says that the first column is the input, and the second
column is the output. sounds reasonable.
the script can also extract specific columns from csv or jsonlines input.
a normal thing to do here (assuming the code works, i could test on
one of their exampels), might be to train a
below is as far as i got.
when i tried running it on xsum, i ran out of gpu ram. i imagine there
are a number of ways to address that.
i'm thinking a simplest approach might be to use a smaller model, even
if it doesn't have the long context support.
another idea is to produce a shorter set of
during adapter training gpu ram usage was under 1700 MB
i actually did this. added adapters to longt5 in my repo branch, and
successfully trained on a csv file consisting of lines equal to "a,b"
. after 400 dataitems, loss dropped to around 0.6, eval loss was 0.0
i don't really know if it's _working_ or just looks like it is or
something, nor how it
Blockchain’s use in healthcare ‘essential’ to protect sensitive data: Zelis CTO
Zelis CTO Kali Durgampudi has called on healthcare providers to utilize
blockchain tech to protect customer data
Ratcheting up the technology, The Candy House fully embraces a science-fiction
conceit, and implies that some of Egan’s minor Goon Squad characters not only
invented social media and seized control of music streaming, but have made good
on a new technology, Own Your Unconscious, which uploads
it's surprising that i did this!
i did not review the implementation structures to see if i missed
anything, once i got it to run
i also did not implement parallelism, tests, nor documentation
-- Forwarded message -
From: Pam Martens
Date: Fri, Jul 22, 2022, 11:02 AM
Subject: Once Daily Digest Email
To:
Wall Street On Parade has posted a new item:
>From the Secret Service to the Fed, Inspectors General Are Enablers to
Corruption at the Agencies they Oversee
-- Forwarded message -
From: Litigation Finance Journal
Date: Fri, Jul 22, 2022, 10:14 AM
Subject: ESG Litigation / Google & Apple Class Action / New Aussie Funder,
and more...
To:
View this email in your browser
Surveillance is pervasive: Yes, you are being watched, even if no one is
looking for you
https://share.newsbreak.com/1h6myng5
Surveillance is pervasive: Yes, you are being watched, even if no one is
looking for you
By Peter Krapp, Professor of Film & Media Studies, University of California,
Brody Larson collates 43 highlights to "THE MATTER OF RUDOLPH W. GIULIANI,
AN ATTORNEY" for reference:
https://drive.google.com/file/d/1vgADU5PlBP6cqJ4pbR2P5JiSgO3bcccG/view?usp=drivesdk
What would Rudy say?
On Fri, Jul 22, 2022, 1:50 PM Gunnar Larson wrote:
> We went from Linda Lacewell to worse
>
> On Fri, Jul 22, 2022, 1:41 PM Gunnar Larson wrote:
>
>>
Dear JP Morgan Chase:
We have yet to hear from you on this important issue.
Gunnar Larson
On Sun, May 1, 2022, 12:21 PM Gunnar Larson wrote:
> May 1, 2022
>
> BY ELECTRONIC MAIL
>
> Investor Relations
> Board of Directors, JPMorgan Chase & Co.
> 277 Park Avenue
> New York, NY 10172-0003
>
While under entrapment, I have been told that being a Cypherpunk is a
terriorist offense.
My reply concerning this matter has been noted as, "Just because you think
the Tooth Ferry has Pink Hair does not mean there is a tooth ferry."
It says nothing about terriorist activity on the Cypherpunk
Ms. James:
We have yet to hear from you on this pressing issue.
Warm regards,
Gunnar Larson
On Sat, Jun 18, 2022, 4:44 PM Gunnar Larson wrote:
> Dear Attorney General Letitia James:
>
> xNY.io - Bank.org, PBC seeks to earn your esteemed approval to engage the
> New York False Claims Act
We went from Linda Lacewell to worse
On Fri, Jul 22, 2022, 1:41 PM Gunnar Larson wrote:
>
Faculty Publications
UC Hastings College of the Law Library
Author: Roger C. Park
Title: The Entrapment Controversy
Source: Minnesota Law Review
Citation: 60 Minn. L. Rev. 163 (1976).
Brody Larson has made 212 highlights:
The investigation behind this keeps getting pushed back on amswers. I am
concerned if the Superintendent cannot be honest about this matter logoc
would signal significant further concerns of her fitness to regulate.
Meanwhile why should all investative bodies in the United States wait for
New
Hello:
Ms. Tomczak:
g...@xny.io is the email on file.
Can you advise why the letter from you today says different?
Gunnar Larson
On Fri, Jul 22, 2022, 2:57 PM wrote:
> Dear Mr. Larson:
>
> You made a FOIL request to Executive Chamber and cc'd the DFS. Our
> response letter is attached.
>
>
Ms. Tomczak:
g...@xny.io is the email on file.
Can you advise why the letter from you today says different?
Gunnar Larson
On Fri, Jul 22, 2022, 2:36 PM Tomczak, Christine (DFS) <
christine.tomc...@dfs.ny.gov> wrote:
> Dear Mr. Larson.
>
>
>
> See attached.
>
>
>
>
>
> Christine M. Tomczak
>
>
Thank you very much.
On Fri, Jul 22, 2022, 3:18 PM Tomczak, Christine (DFS) <
christine.tomc...@dfs.ny.gov> wrote:
> Mr. Larson,
>
>
>
> My apologies, it was simply a typo.
>
>
>
> Christine Tomczak
>
>
>
> *From:* Gunnar Larson
> *Sent:* Friday, July 22, 2022 2:58 PM
> *To:* Tomczak,
" When cypherpunks are called terrorists we will have done our job "
Old cryptoanarchist proverb
Professor Rat - 3 days ago
My beloved country is falling so far behind in Cyber it may never catch-up.
Three examples.
Cypherpunks remain undesignated terrorists
Crypto-muntions remain an
One of the burning issues of the day is how much regulatory arbitrage the USG
may invoke against prime shitcoin competitor, Marxist China. Since it has
nothing but bad choices here
I invoke Napoleon's old
" Never interrupt your enemy when they're making a mistake "
TL/DR full-blown-fascism is
That is a major reason for cypherpunks existence and why we - and others of a
like mind - continue to produce cryptoanarchist infrastructure that negates
this gratuitous intrusion.
We do this or we nothing.
Cypherpunks collapse governments.
Subject: Everyone a remailer: Everyone a Mint:
Good snapshot of this phenomenon - thanks G.
Antitrust was approved of by Mongo and I'd like to maintain his rare good ideas.
Clearly APPLE and GOOGLE are now prime targets for antitrust - and more.
Their CEO class have all earned killing ( My 200 sats )
The state-capitalist that can cut your
On July 22, 2022 2:07:04 PM EDT, Gunnar Larson wrote:
>While under entrapment, I have been told that being a Cypherpunk is a
>terriorist offense.
Gunnar, when you mention being under entrapment, do you mean you are
experiencing pressure of some kind to either break laws or depict yourself as
https://www.mondaq.com/unitedstates/terrorism-homeland-security-defence/1213436/changing-anti-terrorism-act-jurisprudence-increases-litigation-exposure?email_access=on
Is it a contradiction to support assassinations and revolution while also
supporting peaceful petitioning for redress of grievances?
Of course it is for any one individual.
The anarchist is not one isolated middle ' finger ' though. They form '
hand-like ' bands and even ' fists-of-fury '
Phh treble android 12 has been working incredibly on this low end phone for me,
but today misbehaviors began accumulating to the point of notability again.
Although the phone is still incredibly more responsive than on its factory OS.
- popups prevent use of OpenKeychain("ChromeSync" one word,
https://lawandcrime.com/lawsuit/judge-takes-james-comey-andrew-mccabe-peter-strzok-lisa-page-and-kevin-clinesmith-personally-off-the-hook-in-trumps-rico-lawsuit-against-hillary-clinton/
I'd like to add that I'm not really conscious of the phone misbehaviors. If I
knew they were malicious, I would take my devices to authorities or security
consultants or whatnot, but I can't tell if I'm imagining them or not because I
have amnesia around the intermittent things, neurologic
Sorry, Steve " Bane " Bannon found guilty.
This is hardly surprising since video footage exists and will persist of his
brazen attack on Gotham
https://www.youtube.com/watch?v=h9xNy5z8PIA
Russia and Ukraine have reached a deal to resume grain exports through the
Black Sea, in a move that could avert a global food crisis.
https://anarchistnews.org/content/ongoing-sabotage-and-resistance-war-russia-and-ukraine
Reposts etc
Bitcoin Traders in Argentina Restricted From Buying Dollars
It’s an attempt by the central bank to stop the flight of capital and tame
inflation. . . ."
https://decrypt.co/105772/bitcoin-traders-argentina-restricted-buying-dollars
Reposts supplied ' as is '
Eternal shitcoin in the
Some people have to fight for the right to use birth control! Jesus Fucking
Christie!
https://anarchistnews.org/content/part-1-2-birth-control-pioneers-jerk-who-tried-stop-them
Repsots etc
ACID, amnesty and Abortion are mandatory in any global revolution - viva la
revolution
briankrebs
·7h
After reading yet another PR pitch that was basically “blah blah threat blah
threat threat blah blah” I wondered if http://threatthreat.com was taken, b/c
you know it’s hard to come up with good company names for cybering. It was. But
oh ♥ http://threatthreat.com
Between the
Seems strange that Cambridge Analytica is not being debated with Gotham
https://www.reuters.com/legal/litigation/metas-zuckerberg-sandberg-be-deposed-privacy-class-action-lawsuit-2022-07-20/
Maybe even before Bill Gates or the POTUS ( My 200 sats each )
https://stiffs.com/celebrity/markzuckerberg
Any prediction market on any person is also an assassination market on that
person.
85 matches
Mail list logo