From Newsgroup: sci.physics

Hi,

Why does this Lama have a red pyjama.
Oh, its a baby Lama. Its still in the cradle
and needs some training:

RedPajama-Data-v2
https://github.com/togethercomputer/RedPajama-Data

But then Andrej Karpathy recently showed
GPT-2 training on rented GPUs for less
than 100 USD in less then 2 hours.

So where do these grown up Lamas go.
Well Georgi Gerganov prefered C++/C
when he shouted Llama Llama Red Pyjama.

But you also find WebLLM, wrapping the
underlying C++/C GPU interface via the
W3C standard WebGPU / WGSL, with JavaScript:

In-Browser LLM Inference Engine
https://webllm.mlc.ai/

My experience with WebLLM 6 months
ago on an iPad Pro 2024, still a little early
stage performance and robustness.

But hey hardware of AI mobile iGPUs is
still evolving, and AI laptop, AI smartphones
and AI tablets, will soon feature Chinese

hardware such some new Kirin AI in 2027.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly. https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Usual question:

Why implement both pre-emptive threading

AND cooperative tasks/engines?

I had implemented the ISO proposal in formerly Jekejeke
Prolog, you find the ISO proposal here:

ISO/IEC DTR 13211rCo5:2007
Prolog multi-threading support
https://logtalk.org/plstd/threads.pdf

But the ISO proposal doesn't match modern WebGPU APIs,
where your logical threads can live remotely in a dedicated GPU
in the VRAM there, and where you would have launch

parameters that say: Hey please run 4096 compute
shaders for me, that have independet thread state. Using
cooperative multi-tasking as the orchestrator works well.

Bye

Mild Shock schrieb:

Hi,

Why does this Lama have a red pyjama.
Oh, its a baby Lama. Its still in the cradle
and needs some training:

RedPajama-Data-v2
https://github.com/togethercomputer/RedPajama-Data

But then Andrej Karpathy recently showed
GPT-2 training on rented GPUs for less
than 100 USD in less then 2 hours.

So where do these grown up Lamas go.
Well Georgi Gerganov prefered C++/C
when he shouted Llama Llama Red Pyjama.

But you also find WebLLM, wrapping the
underlying C++/C GPU interface via the
W3C standard WebGPU / WGSL, with JavaScript:

In-Browser LLM Inference Engine
https://webllm.mlc.ai/

My experience with WebLLM 6 months
ago on an iPad Pro 2024, still a little early
stage performance and robustness.

But hey hardware of AI mobile iGPUs is
still evolving, and AI laptop, AI smartphones
and AI tablets, will soon feature Chinese

hardware such some new Kirin AI in 2027.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Mostlikely for high performance computing |a la,
the Actor/Erlang model is dead, they might rely
on MPMC (Multiple Producer, Multiple Consumer)

queue entities separate from the threads. The
ISO Prolog multi-threading support had also such
threads. But besides that was also Actor/Erlang

leaning in practice, like SWI, where threads
have some default queues. So an actor is basically
a Thread and Mailbox conflation. While a MPMC queue

is a kind of separate Mailbox, where multiple
"actors" can read from and write from. A kind of
localized Linda Tuple store.

Which Programming language did adopted the
non-Actor pi-calculus model? Right golang
with its channels.

Bye


Mild Shock schrieb:

Hi,

Usual question:

Why implement both pre-emptive threading

AND cooperative tasks/engines?

I had implemented the ISO proposal in formerly Jekejeke
Prolog, you find the ISO proposal here:

ISO/IEC DTR 13211rCo5:2007
Prolog multi-threading support
https://logtalk.org/plstd/threads.pdf

But the ISO proposal doesn't match modern WebGPU APIs,
where your logical threads can live remotely in a dedicated GPU
in the VRAM there, and where you would have launch

parameters that say: Hey please run 4096 compute
shaders for me, that have independet thread state. Using
cooperative multi-tasking as the orchestrator works well.

Bye

Mild Shock schrieb:

Hi,

Why does this Lama have a red pyjama.
Oh, its a baby Lama. Its still in the cradle
and needs some training:

RedPajama-Data-v2
https://github.com/togethercomputer/RedPajama-Data

But then Andrej Karpathy recently showed
GPT-2 training on rented GPUs for less
than 100 USD in less then 2 hours.

So where do these grown up Lamas go.
Well Georgi Gerganov prefered C++/C
when he shouted Llama Llama Red Pyjama.

But you also find WebLLM, wrapping the
underlying C++/C GPU interface via the
W3C standard WebGPU / WGSL, with JavaScript:

In-Browser LLM Inference Engine
https://webllm.mlc.ai/

My experience with WebLLM 6 months
ago on an iPad Pro 2024, still a little early
stage performance and robustness.

But hey hardware of AI mobile iGPUs is
still evolving, and AI laptop, AI smartphones
and AI tablets, will soon feature Chinese

hardware such some new Kirin AI in 2027.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of less than
a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one microsecond.

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, that
the indicators of the above as "positive presence" then is to make
for that the adjustments to the offsets and extents and the shifts
is according to those, otherwise no-ops. Then the idea is that a

As independent logical thread state, that automatically MIMD follows?

Whats the problem to solve then?

Bye

Mild Shock schrieb:

Hi,

Rossy Boy is neither Einstein nor Zweistein.
He is not Einstein since Einstein is already dead:

Albert Einstein (1879 - 1955)
https://de.wikipedia.org/wiki/Albert_Einstein

He is also not Zweistein, since he doesn't
understand concepts such as:

- NVIDIA Volta ff. architecture

Also his hands are small, and his breath stinks,
and he lives in the basement of his mother.

Bye

Mild Shock schrieb:

Hi,

Slowly I start understanding numbnuts like
Rossy Boy who don't understand tech, although
they are from UK and not from a 3rd world

country, and also I start understanding morons
like Micro Penis, who are behind a curtain,
and cannot access a lot of tech.

The same holds for SWI Prologs newest campaign
that probably adresses some poor indians that
have neither 5G nor Macs:

1:38:01 The Kyiv keynote disaster
https://www.youtube.com/watch?v=U8goS6B3BbI

Woa! Real time download of Scala, Closure,
etc.. Whats the magic behind that? Some SWI
point of sale, downloading it via its

keyboard and some telephathy module ?

Bye

Mild Shock schrieb:

Hi,

Ride the snake
He's old and his skin is cold
The west is the best
The west is the best
Get here and we'll do the rest
The blue bus is calling us
The blue bus is calling us
Driver, where you taking us?

Apocalypse Now intro: The Doors, The End {1979}
https://www.youtube.com/watch?v=CIrvSJwwJUE

Bye

Hi,

Again I posted everything here:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

The repo says, same time when I posted
the link first time:

This repository was archived by the
owner on Jul 9, 2026. It is now read-only.

Now a USENET user, who had already entitled
himself for a couple of irrational accusations

towards my side, is asking this question:

Chris M. Thomasson schrieb, Jul 24, 2026

Show an outline of what you
need you compute shader to do?

Bravo, thats a delay of a wooping 15 days.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Hurry Rossy Boy, the blue bus is waiting.
There is a quite a hyperbole from here:

Tesla S1070 in 2008
700 Watts , 1 Terra Flop
SOLVE TOMORROWrCOS PROBLEMS TODAY https://www.azken.com/download/Tesla_DS_S1070_EU.pdf

To here:

Blackwell GPU in 2026
575 Watts, 104.8 Terra Flops ( RTX 5090 )
From Volta To Blackwell https://newsletter.semianalysis.com/p/nvidia-tensor-core-evolution-from-volta-to-blackwell

But somehow the S1070 had already Massively-
Parallel, Many-Core Architecture, and forms
of MIMD, since it had 960 / 240 = 4 cores.

960 scalar processor cores (240 per GPU).
But possibly more resticted inside work
groups, than later NVIDIA Volta ff

architecture with independent thread state.

Bye

Disclaimer: The above is only a very rough
RTX 5090 spec. Its doesn't say what value
format and what vector/matrics ops were

used. Also energy consumption may vary.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of less than
a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one microsecond.

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, that
the indicators of the above as "positive presence" then is to make
for that the adjustments to the offsets and extents and the shifts
is according to those, otherwise no-ops. Then the idea is that a

As independent logical thread state, that automatically MIMD follows?

Whats the problem to solve then?

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

So what does NUM_SHADERS = 4096 shaders mean here?

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

Its only the number of logical threads.

CUDArao TEChNOLOGY UNLOCkS ThE POWER OF TESLA MANY-CORE PROCESSORS
The CUDA C compiler simplifies many-core programming
by enabling code development in a high-level language
and optimizing code to run on systems without knowledge of
how many cores are in the hardware.

CUDA applications automatically take advantage of more
cores or fewer cores in a system, so they can scale from
entry-level notebook GPUs to high end GPUs in technical
workstations and further into racks of GPUs in data
centers. This allows developers to

rCLcode oncerCY and deploy on a range of systems, as well as
scale forward in time as future GPUs deliver more
performance per watt and more cores per processor. The benefit
for software users is the opportunity to boost computing
performance simply by adding GPUs or using their

existing GPUs in new ways.
https://www.azken.com/download/Tesla_DS_S1070_EU.pdf

Bye

Mild Shock schrieb:

Hi,

Hurry Rossy Boy, the blue bus is waiting.
There is a quite a hyperbole from here:

Tesla S1070 in 2008
700 Watts , 1 Terra Flop
SOLVE TOMORROWrCOS PROBLEMS TODAY https://www.azken.com/download/Tesla_DS_S1070_EU.pdf

To here:

Blackwell GPU in 2026
575 Watts, 104.8 Terra Flops ( RTX 5090 )
From Volta To Blackwell https://newsletter.semianalysis.com/p/nvidia-tensor-core-evolution-from-volta-to-blackwell

But somehow the S1070 had already Massively-
Parallel, Many-Core Architecture, and forms
of MIMD, since it had 960 / 240 = 4 cores.

960 scalar processor cores (240 per GPU).
But possibly more resticted inside work
groups, than later NVIDIA Volta ff

architecture with independent thread state.

Bye

Disclaimer: The above is only a very rough
RTX 5090 spec. Its doesn't say what value
format and what vector/matrics ops were

used. Also energy consumption may vary.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of less than >> -a> a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one microsecond.

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, that
the indicators of the above as "positive presence" then is to make
for that the adjustments to the offsets and extents and the shifts
is according to those, otherwise no-ops. Then the idea is that a

As independent logical thread state, that automatically MIMD follows?

Whats the problem to solve then?

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Because of this parallelism you anyway
need to forget about any arithmetization
of product FSA (finite-state automata).

Just forget it. What modern GPU provide
is a kind of hirarchical viewpoint. You
can have barriers in groups etc..

So you can exercise control over your
mongolian horde of logical threads in
a kind of multilevel schema.

Have Fun!

Bye

Mild Shock schrieb:

Hi,

So what does NUM_SHADERS = 4096 shaders mean here?

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

Its only the number of logical threads.

CUDArao TEChNOLOGY UNLOCkS ThE POWER OF TESLA MANY-CORE PROCESSORS
The CUDA C compiler simplifies-a many-core programming
by enabling code development in a high-level language
and optimizing code to run on systems without knowledge of
how many cores are in the hardware.

CUDA applications automatically take advantage of more
cores or fewer cores in a system, so they can scale from
entry-level notebook GPUs to high end GPUs in technical
workstations-a and further into racks of GPUs in data
centers. This allows developers to

rCLcode oncerCY and deploy on a range of systems, as well as
scale forward in time as future GPUs deliver more
performance per watt and more cores per processor. The benefit
for software users is the opportunity to boost computing
performance simply by adding GPUs or using their

existing GPUs in new ways. https://www.azken.com/download/Tesla_DS_S1070_EU.pdf

Bye

Mild Shock schrieb:

Hi,

Hurry Rossy Boy, the blue bus is waiting.
There is a quite a hyperbole from here:

Tesla S1070 in 2008
700 Watts , 1 Terra Flop
SOLVE TOMORROWrCOS PROBLEMS TODAY
https://www.azken.com/download/Tesla_DS_S1070_EU.pdf

To here:

Blackwell GPU in 2026
575 Watts, 104.8 Terra Flops ( RTX 5090 )
-aFrom Volta To Blackwell
https://newsletter.semianalysis.com/p/nvidia-tensor-core-evolution-from-volta-to-blackwell


But somehow the S1070 had already Massively-
Parallel, Many-Core Architecture, and forms
of MIMD, since it had 960 / 240 = 4 cores.

960 scalar processor cores (240 per GPU).
But possibly more resticted inside work
groups, than later NVIDIA Volta ff

architecture with independent thread state.

Bye

Disclaimer: The above is only a very rough
RTX 5090 spec. Its doesn't say what value
format and what vector/matrics ops were

used. Also energy consumption may vary.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of less

than

a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one microsecond.

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, that
the indicators of the above as "positive presence" then is to make
for that the adjustments to the offsets and extents and the shifts
is according to those, otherwise no-ops. Then the idea is that a

As independent logical thread state, that automatically MIMD follows?

Whats the problem to solve then?

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

This was archived on Jul 9, 2026:

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

Still, Jul 29, Rossy Boy halucinates accusations:

Ross Finlayson schrieb:

.. bla bla goto bla bla ..

Stupid gangster: teamsters are a union.

In the trades, not the steals, ....

Woa! Thats now 20 days of brain desease,
and not understanding the meaning and implications.
Even not understand pi-WAM has Hack VM backend.

But its all opensource. Bravo Rossy Boy, you are
champion in brainlessness and lazyness of
a idiot usenet troll.

Bye

Mild Shock schrieb:

Hi,

If any of you guys do not understand what
is meant by or what the implications are:

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

Well I wouldn't care less. There are two
outcomes for numb nuts:

- Ignoramus: They don't understand it, but
-a they will understand it before they die.

- Ignorabimus: They don't understand it, and
-a will never understand it, and they die.

So who cares, its not my problem, you people
are stupid as fuck, and slow as fuck...

Bye

Mild Shock schrieb:

Hi,

Micro penis brain is in constant hiatus.
He can even not detect a trope.

LoL

Bye

Lane W schrieb:

Mild Shock wrote:

Hi,

My mother is worried that I fucked Lane W.
aka Micro Penis mother 24 hours straight.
She was screaming, basically singing all

the arias from operas that Luciano Pavarotti
usually sings. You Lane W. aka Micro Penis
should have heard it, since you

live in the basement of your mothers house.

No, actually remarkably, I don't. According to google I live 433

miles away from her.

Strike!

See, what i said about you was spot on.

What you said about me was generic and incorrect.

You really suck, man.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Rossy Boys tears could cool a data center,
he thinks there exists no literature about
serial algorithms of parallel stuff, and

he also thinks normal forms lead to optimizing
something. LoL, what a utter bullshit. I did
alreay a serial implementation of a parallel

simulation of my pi-WAM. Just lookup the literature
about pi-calulus. I published it a few days ago,
its part of 2.2.4 released already:

Parallel -C-WAM: An Interleaved Synchronous Emulator https://medium.com/2989/0196089e143a

Whats your point, Rossy Boy? Except you post pretend
nonsense not knowing what you are doing?

Bye

Ross Finlayson schrieb:

No, troll, these are serial algorithms their optimized forms.

Normal sorts of forms, ....

Yeah, everybody already figured out "interpreters" and
"programs" and "spawning".

Go spawn yourself.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of less than
a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one microsecond.

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, that
the indicators of the above as "positive presence" then is to make
for that the adjustments to the offsets and extents and the shifts
is according to those, otherwise no-ops. Then the idea is that a

As independent logical thread state, that automatically MIMD follows?

Whats the problem to solve then?

Bye

Mild Shock schrieb:

Hi,

Rossy Boy is neither Einstein nor Zweistein.
He is not Einstein since Einstein is already dead:

Albert Einstein (1879 - 1955)
https://de.wikipedia.org/wiki/Albert_Einstein

He is also not Zweistein, since he doesn't
understand concepts such as:

- NVIDIA Volta ff. architecture

Also his hands are small, and his breath stinks,
and he lives in the basement of his mother.

Bye

Mild Shock schrieb:

Hi,

Slowly I start understanding numbnuts like
Rossy Boy who don't understand tech, although
they are from UK and not from a 3rd world

country, and also I start understanding morons
like Micro Penis, who are behind a curtain,
and cannot access a lot of tech.

The same holds for SWI Prologs newest campaign
that probably adresses some poor indians that
have neither 5G nor Macs:

1:38:01 The Kyiv keynote disaster
https://www.youtube.com/watch?v=U8goS6B3BbI

Woa! Real time download of Scala, Closure,
etc.. Whats the magic behind that? Some SWI
point of sale, downloading it via its

keyboard and some telephathy module ?

Bye

Mild Shock schrieb:

Hi,

Ride the snake
He's old and his skin is cold
The west is the best
The west is the best
Get here and we'll do the rest
The blue bus is calling us
The blue bus is calling us
Driver, where you taking us?

Apocalypse Now intro: The Doors, The End {1979}
https://www.youtube.com/watch?v=CIrvSJwwJUE

Bye

Hi,

Again I posted everything here:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

The repo says, same time when I posted
the link first time:

This repository was archived by the
owner on Jul 9, 2026. It is now read-only.

Now a USENET user, who had already entitled
himself for a couple of irrational accusations

towards my side, is asking this question:

Chris M. Thomasson schrieb, Jul 24, 2026

Show an outline of what you
need you compute shader to do?

Bravo, thats a delay of a wooping 15 days.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

I don't use Rust, you are crazy. First of
all the parallel simulator is 100% written
in Prolog, should also run in ISO Prolog,

enhanced by a library(lists). Second I only
mentioned that WebGPU / WGSL, the language
there has a Rust inspired language.

Its not Rust. Whats wrong with you? Why do
you adress your weariness of life to me.
I am neither thief, nor can I help you

with your frustration, and histeric outbursts.
Maybe just be a man and jump off a bridge, idiot.
Or tame your frustration, usenet is not for

you alone, your stupid asshole.

Bye

Ross Finlayson schrieb:

https://www.theregister.com/databases/2026/07/29/after-rewriting-sqlite-in-rust-turso-turns-its-sights-on-postgres/5279835

I don't much care about Rust.

.. gibberish ..

Thief.

Mild Shock schrieb:

Hi,

Rossy Boys tears could cool a data center,
he thinks there exists no literature about
serial algorithms of parallel stuff, and

he also thinks normal forms lead to optimizing
something. LoL, what a utter bullshit. I did
alreay a serial implementation of a parallel

simulation of my pi-WAM. Just lookup the literature
about pi-calulus. I published it a few days ago,
its part of 2.2.4 released already:

Parallel -C-WAM: An Interleaved Synchronous Emulator https://medium.com/2989/0196089e143a

Whats your point, Rossy Boy? Except you post pretend
nonsense not knowing what you are doing?

Bye

Ross Finlayson schrieb:

No, troll, these are serial algorithms their optimized forms.

Normal sorts of forms, ....

Yeah, everybody already figured out "interpreters" and
"programs" and "spawning".

Go spawn yourself.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of less than >> -a> a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one microsecond.

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, that
the indicators of the above as "positive presence" then is to make
for that the adjustments to the offsets and extents and the shifts
is according to those, otherwise no-ops. Then the idea is that a

As independent logical thread state, that automatically MIMD follows?

Whats the problem to solve then?

Bye

Mild Shock schrieb:

Hi,

Rossy Boy is neither Einstein nor Zweistein.
He is not Einstein since Einstein is already dead:

Albert Einstein (1879 - 1955)
https://de.wikipedia.org/wiki/Albert_Einstein

He is also not Zweistein, since he doesn't
understand concepts such as:

- NVIDIA Volta ff. architecture

Also his hands are small, and his breath stinks,
and he lives in the basement of his mother.

Bye

Mild Shock schrieb:

Hi,

Slowly I start understanding numbnuts like
Rossy Boy who don't understand tech, although
they are from UK and not from a 3rd world

country, and also I start understanding morons
like Micro Penis, who are behind a curtain,
and cannot access a lot of tech.

The same holds for SWI Prologs newest campaign
that probably adresses some poor indians that
have neither 5G nor Macs:

1:38:01 The Kyiv keynote disaster
https://www.youtube.com/watch?v=U8goS6B3BbI

Woa! Real time download of Scala, Closure,
etc.. Whats the magic behind that? Some SWI
point of sale, downloading it via its

keyboard and some telephathy module ?

Bye

Mild Shock schrieb:

Hi,

Ride the snake
He's old and his skin is cold
The west is the best
The west is the best
Get here and we'll do the rest
The blue bus is calling us
The blue bus is calling us
Driver, where you taking us?

Apocalypse Now intro: The Doors, The End {1979}
https://www.youtube.com/watch?v=CIrvSJwwJUE

Bye

Hi,

Again I posted everything here:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

The repo says, same time when I posted
the link first time:

This repository was archived by the
owner on Jul 9, 2026. It is now read-only.

Now a USENET user, who had already entitled
himself for a couple of irrational accusations

towards my side, is asking this question:

Chris M. Thomasson schrieb, Jul 24, 2026

Show an outline of what you
need you compute shader to do?

Bravo, thats a delay of a wooping 15 days.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Who exactly is the thief? Does this person
have stats in the Rogue class in dungeons
and dragons?

The conspiracy theory of a stealing of Torso VDBE,
by Rossy Boy, is probably a result of complete
ignorance of the Hack ecosystem.

Hack is a very popular computer science project,
with a couple of subprojects in hardware and
software. It goes also by the name Nand to Tetris,

and is programming language agnositic. You can do
Hack experiments in any programming language, be
it BASIC, ADA or Rust. Nobody cares.

The gist are projects like here, first to
educate yourself about Hack:

https://www.nand2tetris.org/course

And then to use Hack in different contexts:

https://www.nand2tetris.org/copy-of-talks

For didactic purposes, I used Hack for my WebGPU
experiment. I didn't even take a look at Torso
VDBE, why should I? Hack is nicely documented,

has even a book, and fusing the two 16-bit
instruction types A and D, into a single 32-bit
instruction stream, is nowhere patented.

Bye

Mild Shock schrieb:

Hi,

I don't use Rust, you are crazy. First of
all the parallel simulator is 100% written
in Prolog, should also run in ISO Prolog,

enhanced by a library(lists). Second I only
mentioned that WebGPU / WGSL, the language
there has a Rust inspired language.

Its not Rust. Whats wrong with you? Why do
you adress your weariness of life to me.
I am neither thief, nor can I help you

with your frustration, and histeric outbursts.
Maybe just be a man and jump off a bridge, idiot.
Or tame your frustration, usenet is not for

you alone, your stupid asshole.

Bye

Ross Finlayson schrieb:

https://www.theregister.com/databases/2026/07/29/after-rewriting-sqlite-in-rust-turso-turns-its-sights-on-postgres/5279835

I don't much care about Rust.

.. gibberish ..

Thief.

Mild Shock schrieb:

Hi,

Rossy Boys tears could cool a data center,
he thinks there exists no literature about
serial algorithms of parallel stuff, and

he also thinks normal forms lead to optimizing
something. LoL, what a utter bullshit. I did
alreay a serial implementation of a parallel

simulation of my pi-WAM. Just lookup the literature
about pi-calulus. I published it a few days ago,
its part of 2.2.4 released already:

Parallel -C-WAM: An Interleaved Synchronous Emulator
https://medium.com/2989/0196089e143a

Whats your point, Rossy Boy? Except you post pretend
nonsense not knowing what you are doing?

Bye

Ross Finlayson schrieb:

No, troll, these are serial algorithms their optimized forms.

Normal sorts of forms, ....


Yeah, everybody already figured out "interpreters" and
"programs" and "spawning".

Go spawn yourself.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of less

than

a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one microsecond.

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, that
the indicators of the above as "positive presence" then is to make
for that the adjustments to the offsets and extents and the shifts
is according to those, otherwise no-ops. Then the idea is that a

As independent logical thread state, that automatically MIMD follows?

Whats the problem to solve then?

Bye

Mild Shock schrieb:

Hi,

Rossy Boy is neither Einstein nor Zweistein.
He is not Einstein since Einstein is already dead:

Albert Einstein (1879 - 1955)
https://de.wikipedia.org/wiki/Albert_Einstein

He is also not Zweistein, since he doesn't
understand concepts such as:

- NVIDIA Volta ff. architecture

Also his hands are small, and his breath stinks,
and he lives in the basement of his mother.

Bye

Mild Shock schrieb:

Hi,

Slowly I start understanding numbnuts like
Rossy Boy who don't understand tech, although
they are from UK and not from a 3rd world

country, and also I start understanding morons
like Micro Penis, who are behind a curtain,
and cannot access a lot of tech.

The same holds for SWI Prologs newest campaign
that probably adresses some poor indians that
have neither 5G nor Macs:

1:38:01 The Kyiv keynote disaster
https://www.youtube.com/watch?v=U8goS6B3BbI

Woa! Real time download of Scala, Closure,
etc.. Whats the magic behind that? Some SWI
point of sale, downloading it via its

keyboard and some telephathy module ?

Bye

Mild Shock schrieb:

Hi,

Ride the snake
He's old and his skin is cold
The west is the best
The west is the best
Get here and we'll do the rest
The blue bus is calling us
The blue bus is calling us
Driver, where you taking us?

Apocalypse Now intro: The Doors, The End {1979}
https://www.youtube.com/watch?v=CIrvSJwwJUE

Bye

Hi,

Again I posted everything here:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

The repo says, same time when I posted
the link first time:

This repository was archived by the
owner on Jul 9, 2026. It is now read-only.

Now a USENET user, who had already entitled
himself for a couple of irrational accusations

towards my side, is asking this question:

Chris M. Thomasson schrieb, Jul 24, 2026

Show an outline of what you
need you compute shader to do?

Bravo, thats a delay of a wooping 15 days.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

This seems to be a funny Q16.16 experiment.
It shows that an integerish Hack can do
floatish stuff, by using binary fixpoint:

Raytracing on the Hack computer
2021/06/13 - im alex
https://blog.alexqua.ch/posts/from-nand-to-raytracer/

That it uses Rust is arbitrary. Feel free
to do it in C, C++, FORTRAN or Java. I guess
these languages all have basic arithmethic,

right? Maybe not a long jump always?

Bye

Mild Shock schrieb:

Hi,

Who exactly is the thief? Does this person
have stats in the Rogue class in dungeons
and dragons?

The conspiracy theory of a stealing of Torso VDBE,
by Rossy Boy, is probably a result of complete
ignorance of the Hack ecosystem.

Hack is a very popular computer science project,
with a couple of subprojects in hardware and
software. It goes also by the name Nand to Tetris,

and is programming language agnositic. You can do
Hack experiments in any programming language, be
it BASIC, ADA or Rust. Nobody cares.

The gist are projects like here, first to
educate yourself about Hack:

https://www.nand2tetris.org/course

And then to use Hack in different contexts:

https://www.nand2tetris.org/copy-of-talks

For didactic purposes, I used Hack for my WebGPU
experiment. I didn't even take a look at Torso
VDBE, why should I? Hack is nicely documented,

has even a book, and fusing the two 16-bit
instruction types A and D, into a single 32-bit
instruction stream, is nowhere patented.

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

It explicitly states:

GPUs have evolved form lock-step to independent
thread scheduling. This made it possible to port the
Hack VM variant, that forms the basis for our -C-WAM,
to WebGPU computer shaders. Using NUM_SHADERS = 4096
we could produce 11.4 Giga Lips on a
Ryzen AI 7 350 w/ Radeon 860M.

Here is one for 605.- CHF:

ASUS Vivobook 14
Prozessortyp AMD Ryzen AI 7 350
Grafikkarten Modell Radeon 860M https://www.digitec.ch/de/s1/product/asus-vivobook-14-14-512-gb-16-gb-ch-amd-ryzen-ai-7-350-notebook-54670603

Bye

Mantra Mahonov schrieb:

Mild Shock wrote:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

post the link to the retailer, see what he sells for what money, idiot

The AMD Ryzen AI Halo PC is available for order exclusively through Micro Center in the United States. It is priced at $3,999.99

Mild Shock schrieb:

Hi,

This was archived on Jul 9, 2026:

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

Still, Jul 29, Rossy Boy halucinates accusations:

Ross Finlayson schrieb:

.. bla bla goto bla bla ..

Stupid gangster:-a teamsters are a union.

In the trades, not the steals, ....

Woa! Thats now 20 days of brain desease,
and not understanding the meaning and implications.
Even not understand pi-WAM has Hack VM backend.

But its all opensource. Bravo Rossy Boy, you are
champion in brainlessness and lazyness of
a idiot usenet troll.

Bye

Mild Shock schrieb:

Hi,

If any of you guys do not understand what
is meant by or what the implications are:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

Well I wouldn't care less. There are two
outcomes for numb nuts:

- Ignoramus: They don't understand it, but
-a-a they will understand it before they die.

- Ignorabimus: They don't understand it, and
-a-a will never understand it, and they die.

So who cares, its not my problem, you people
are stupid as fuck, and slow as fuck...

Bye

Mild Shock schrieb:

Hi,

Micro penis brain is in constant hiatus.
He can even not detect a trope.

LoL

Bye

Lane W schrieb:

Mild Shock wrote:

Hi,

My mother is worried that I fucked Lane W.
aka Micro Penis mother 24 hours straight.
She was screaming, basically singing all

the arias from operas that Luciano Pavarotti
usually sings. You Lane W. aka Micro Penis
should have heard it, since you

live in the basement of your mothers house.

No, actually remarkably, I don't. According to google I live 433

miles away from her.

Strike!

See, what i said about you was spot on.

What you said about me was generic and incorrect.

You really suck, man.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Woa! Thats a very sad and non fitting statement:

"This was before I was indoctrinated into
ISO Prolog and the ways of monotonic logic
programming. Shen Prolog has many semantic
and syntactic limitations that Scryer Prolog
does not. Also, I now know constraints are a
much better, purer solution to the problems
mode declarations were meant to address" https://github.com/mthom/scryer-prolog/issues/3410#issuecomment-5030471183

Ok, here is the summer challenge, thats the easy one:

SQL --> Prolog --> WAN

Here come two variations, slightly mindboggling maybe?

SQL --> AST --> VDBE

SQL --> Prolog+Modes --> -C-WAM

Bye

BTW: What is VDBE? Some abstract machine, that can
be used to run SQL, following some ideas here:

Database Co-Design With Asynchronous I/O https://penberg.org/papers/penberg-edgesys24.pdf

Or to run Doom:

Doom on the Turso VDBE
https://github.com/tursodatabase/turso-vdbe-doom-example

What if we would run Doom with -C-WAM, on a GPU,
using multiple shaders. We could add some ray tracing.

Mild Shock schrieb:

Hi,

This seems to be a funny Q16.16 experiment.
It shows that an integerish Hack can do
floatish stuff, by using binary fixpoint:

Raytracing on the Hack computer
2021/06/13 - im alex
https://blog.alexqua.ch/posts/from-nand-to-raytracer/

That it uses Rust is arbitrary. Feel free
to do it in C, C++, FORTRAN or Java. I guess
these languages all have basic arithmethic,

right? Maybe not a long jump always?

Bye

Mild Shock schrieb:

Hi,

Who exactly is the thief? Does this person
have stats in the Rogue class in dungeons
and dragons?

The conspiracy theory of a stealing of Torso VDBE,
by Rossy Boy, is probably a result of complete
ignorance of the Hack ecosystem.

Hack is a very popular computer science project,
with a couple of subprojects in hardware and
software. It goes also by the name Nand to Tetris,

and is programming language agnositic. You can do
Hack experiments in any programming language, be
it BASIC, ADA or Rust. Nobody cares.

The gist are projects like here, first to
educate yourself about Hack:

https://www.nand2tetris.org/course

And then to use Hack in different contexts:

https://www.nand2tetris.org/copy-of-talks

For didactic purposes, I used Hack for my WebGPU
experiment. I didn't even take a look at Torso
VDBE, why should I? Hack is nicely documented,

has even a book, and fusing the two 16-bit
instruction types A and D, into a single 32-bit
instruction stream, is nowhere patented.

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Just downloading some other person's code

I didn't do that, I wrote Hack VM for pi-WAM
from scratch, over the last 4 weeks. I came
back from holidays on end of June 2026, and now

we have end of July 2026. But its only possible
because the instruction set is very smal, like
ca. 8 functions and ca. 8 modes and ca. 8 conditions,

so its ca. 8 x 8 x 8 = 512 opcodes, each has an
A parameter and a D parameter simultaneously.
It has currently the following CPU backends:

- Now supports interleaved synchronous emulation.
- Now supports warp parallelism via Java platform threads.
- Now supports warp parallelism via Python system threads.
- Now supports warp parallelism via JavaScript worker threads.
- Note: For Python free threads are not yet fully tested.
- Note: For JavaScript web workers are not yet fully tested.

https://www.dogelog.ch/typtab/doclet/book/14_install/05_notes22/110_224.html

But frankly I came to encounter Hack not from
the usual university curriculum web resources,
but indirectly through a post about a Prolog

emulation of Hack, using constrained horn clauses (CHC):

Verifying Nand2Tetris Assembly
https://www.philipzucker.com/nand2tetris-chc/

The binary encoding is currently that the functions,
modes and conditions eat up a nibble (4-bit), in
total 12-bit, which I use then 10-bit for A parameter

and 10-bit for D parameter. I used AI freemium, Codex
by ChatGPT from within IntelliJ to do some fragment
code translations automatically from Java to JavaScript

or from JavaScript to Python.

Have Fun!

Bye

Mild Shock schrieb:

Hi,

Who exactly is the thief? Does this person
have stats in the Rogue class in dungeons
and dragons?

The conspiracy theory of a stealing of Torso VDBE,
by Rossy Boy, is probably a result of complete
ignorance of the Hack ecosystem.

Hack is a very popular computer science project,
with a couple of subprojects in hardware and
software. It goes also by the name Nand to Tetris,

and is programming language agnositic. You can do
Hack experiments in any programming language, be
it BASIC, ADA or Rust. Nobody cares.

The gist are projects like here, first to
educate yourself about Hack:

https://www.nand2tetris.org/course

And then to use Hack in different contexts:

https://www.nand2tetris.org/copy-of-talks

For didactic purposes, I used Hack for my WebGPU
experiment. I didn't even take a look at Torso
VDBE, why should I? Hack is nicely documented,

has even a book, and fusing the two 16-bit
instruction types A and D, into a single 32-bit
instruction stream, is nowhere patented.

Bye

Mild Shock schrieb:

Hi,

I don't use Rust, you are crazy. First of
all the parallel simulator is 100% written
in Prolog, should also run in ISO Prolog,

enhanced by a library(lists). Second I only
mentioned that WebGPU / WGSL, the language
there has a Rust inspired language.

Its not Rust. Whats wrong with you? Why do
you adress your weariness of life to me.
I am neither thief, nor can I help you

with your frustration, and histeric outbursts.
Maybe just be a man and jump off a bridge, idiot.
Or tame your frustration, usenet is not for

you alone, your stupid asshole.

Bye

Ross Finlayson schrieb:

https://www.theregister.com/databases/2026/07/29/after-rewriting-sqlite-in-rust-turso-turns-its-sights-on-postgres/5279835

I don't much care about Rust.

.. gibberish ..

Thief.

Mild Shock schrieb:

Hi,

Rossy Boys tears could cool a data center,
he thinks there exists no literature about
serial algorithms of parallel stuff, and

he also thinks normal forms lead to optimizing
something. LoL, what a utter bullshit. I did
alreay a serial implementation of a parallel

simulation of my pi-WAM. Just lookup the literature
about pi-calulus. I published it a few days ago,
its part of 2.2.4 released already:

Parallel -C-WAM: An Interleaved Synchronous Emulator
https://medium.com/2989/0196089e143a

Whats your point, Rossy Boy? Except you post pretend
nonsense not knowing what you are doing?

Bye

Ross Finlayson schrieb:

No, troll, these are serial algorithms their optimized forms.

Normal sorts of forms, ....


Yeah, everybody already figured out "interpreters" and
"programs" and "spawning".

Go spawn yourself.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of less >>>> than
a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one microsecond. >>>>

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, that >>>> -a> the indicators of the above as "positive presence" then is to make >>>> -a> for that the adjustments to the offsets and extents and the shifts >>>> -a> is according to those, otherwise no-ops. Then the idea is that a

As independent logical thread state, that automatically MIMD follows?

Whats the problem to solve then?

Bye

Mild Shock schrieb:

Hi,

Rossy Boy is neither Einstein nor Zweistein.
He is not Einstein since Einstein is already dead:

Albert Einstein (1879 - 1955)
https://de.wikipedia.org/wiki/Albert_Einstein

He is also not Zweistein, since he doesn't
understand concepts such as:

- NVIDIA Volta ff. architecture

Also his hands are small, and his breath stinks,
and he lives in the basement of his mother.

Bye

Mild Shock schrieb:

Hi,

Slowly I start understanding numbnuts like
Rossy Boy who don't understand tech, although
they are from UK and not from a 3rd world

country, and also I start understanding morons
like Micro Penis, who are behind a curtain,
and cannot access a lot of tech.

The same holds for SWI Prologs newest campaign
that probably adresses some poor indians that
have neither 5G nor Macs:

1:38:01 The Kyiv keynote disaster
https://www.youtube.com/watch?v=U8goS6B3BbI

Woa! Real time download of Scala, Closure,
etc.. Whats the magic behind that? Some SWI
point of sale, downloading it via its

keyboard and some telephathy module ?

Bye

Mild Shock schrieb:

Hi,

Ride the snake
He's old and his skin is cold
The west is the best
The west is the best
Get here and we'll do the rest
The blue bus is calling us
The blue bus is calling us
Driver, where you taking us?

Apocalypse Now intro: The Doors, The End {1979}
https://www.youtube.com/watch?v=CIrvSJwwJUE

Bye

Hi,

Again I posted everything here:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

The repo says, same time when I posted
the link first time:

This repository was archived by the
owner on Jul 9, 2026. It is now read-only.

Now a USENET user, who had already entitled
himself for a couple of irrational accusations

towards my side, is asking this question:

Chris M. Thomasson schrieb, Jul 24, 2026

Show an outline of what you
need you compute shader to do?

Bravo, thats a delay of a wooping 15 days.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

The op-codes are all uniform, have the
same sub fields. Already Z-80 CPU differs here.
Another difference to a Z-80 CPU is that

their instruction stream was 8-bit, instructions
can 1, 2, 3 or 4 byte long. On the other
hand in my Hack VM all instructions are

one 32-bit chunk. The porting of a first
prototype that I already had, to WebGPU / WGSL
only took like 1-2 hours. The execution

of Hack VM is very simple, version 1.0,
for a single shader:

fn run() {
var pc : i32 = 0;
var accu : i32 = 0;
while (pc < i32(arrayLength(&code))) {
var instr : i32 = code[pc];
pc += 1;
var value : i32 = run_get(instr);
accu = run_fun(instr, accu, value);
run_set(instr, accu);
pc += run_jump(instr, accu);
}
}

https://github.com/Jean-Luc-Picard-2021/gigabudget/blob/b8946e891be774c40522267ab17062d32b023e7a/course/example63/boot.mjs#L176-L187

I first though this will be perfect for
SIMD. Until I learnt that modern GPUs have
anyway MIMD. Hell Yeah, thats much better!

Bye

Mild Shock schrieb:

Hi,

Just downloading some other person's code

I didn't do that, I wrote Hack VM for pi-WAM
from scratch, over the last 4 weeks. I came
back from holidays on end of June 2026, and now

we have end of July 2026. But its only possible
because the instruction set is very smal, like
ca. 8 functions and ca. 8 modes and ca. 8 conditions,

so its ca. 8 x 8 x 8 = 512 opcodes, each has an
A parameter and a D parameter simultaneously.
It has currently the following CPU backends:

-a- Now supports interleaved synchronous emulation.
-a- Now supports warp parallelism via Java platform threads.
-a- Now supports warp parallelism via Python system threads.
-a- Now supports warp parallelism via JavaScript worker threads.
-a- Note: For Python free threads are not yet fully tested.
-a- Note: For JavaScript web workers are not yet fully tested.

https://www.dogelog.ch/typtab/doclet/book/14_install/05_notes22/110_224.html

But frankly I came to encounter Hack not from
the usual university curriculum web resources,
but indirectly through a post about a Prolog

emulation of Hack, using constrained horn clauses (CHC):

Verifying Nand2Tetris Assembly
https://www.philipzucker.com/nand2tetris-chc/

The binary encoding is currently that the functions,
modes and conditions eat up a nibble (4-bit), in
total 12-bit, which I use then 10-bit for A parameter

and 10-bit for D parameter. I used AI freemium, Codex
by ChatGPT from within IntelliJ to do some fragment
code translations automatically from Java to JavaScript

or from JavaScript to Python.

Have Fun!

Bye

Mild Shock schrieb:

Hi,

Who exactly is the thief? Does this person
have stats in the Rogue class in dungeons
and dragons?

The conspiracy theory of a stealing of Torso VDBE,
by Rossy Boy, is probably a result of complete
ignorance of the Hack ecosystem.

Hack is a very popular computer science project,
with a couple of subprojects in hardware and
software. It goes also by the name Nand to Tetris,

and is programming language agnositic. You can do
Hack experiments in any programming language, be
it BASIC, ADA or Rust. Nobody cares.

The gist are projects like here, first to
educate yourself about Hack:

https://www.nand2tetris.org/course

And then to use Hack in different contexts:

https://www.nand2tetris.org/copy-of-talks

For didactic purposes, I used Hack for my WebGPU
experiment. I didn't even take a look at Torso
VDBE, why should I? Hack is nicely documented,

has even a book, and fusing the two 16-bit
instruction types A and D, into a single 32-bit
instruction stream, is nowhere patented.

Bye

Mild Shock schrieb:

Hi,

I don't use Rust, you are crazy. First of
all the parallel simulator is 100% written
in Prolog, should also run in ISO Prolog,

enhanced by a library(lists). Second I only
mentioned that WebGPU / WGSL, the language
there has a Rust inspired language.

Its not Rust. Whats wrong with you? Why do
you adress your weariness of life to me.
I am neither thief, nor can I help you

with your frustration, and histeric outbursts.
Maybe just be a man and jump off a bridge, idiot.
Or tame your frustration, usenet is not for

you alone, your stupid asshole.

Bye

Ross Finlayson schrieb:

https://www.theregister.com/databases/2026/07/29/after-rewriting-sqlite-in-rust-turso-turns-its-sights-on-postgres/5279835

I don't much care about Rust.

.. gibberish ..

Thief.

Mild Shock schrieb:

Hi,

Rossy Boys tears could cool a data center,
he thinks there exists no literature about
serial algorithms of parallel stuff, and

he also thinks normal forms lead to optimizing
something. LoL, what a utter bullshit. I did
alreay a serial implementation of a parallel

simulation of my pi-WAM. Just lookup the literature
about pi-calulus. I published it a few days ago,
its part of 2.2.4 released already:

Parallel -C-WAM: An Interleaved Synchronous Emulator
https://medium.com/2989/0196089e143a

Whats your point, Rossy Boy? Except you post pretend
nonsense not knowing what you are doing?

Bye

Ross Finlayson schrieb:

No, troll, these are serial algorithms their optimized forms.

Normal sorts of forms, ....


Yeah, everybody already figured out "interpreters" and
"programs" and "spawning".

Go spawn yourself.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of

less than

a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one microsecond. >>>>>

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, that >>>>> -a> the indicators of the above as "positive presence" then is to make >>>>> -a> for that the adjustments to the offsets and extents and the shifts >>>>> -a> is according to those, otherwise no-ops. Then the idea is that a >>>>>

As independent logical thread state, that automatically MIMD follows? >>>>>
Whats the problem to solve then?

Bye

Mild Shock schrieb:

Hi,

Rossy Boy is neither Einstein nor Zweistein.
He is not Einstein since Einstein is already dead:

Albert Einstein (1879 - 1955)
https://de.wikipedia.org/wiki/Albert_Einstein

He is also not Zweistein, since he doesn't
understand concepts such as:

- NVIDIA Volta ff. architecture

Also his hands are small, and his breath stinks,
and he lives in the basement of his mother.

Bye

Mild Shock schrieb:

Hi,

Slowly I start understanding numbnuts like
Rossy Boy who don't understand tech, although
they are from UK and not from a 3rd world

country, and also I start understanding morons
like Micro Penis, who are behind a curtain,
and cannot access a lot of tech.

The same holds for SWI Prologs newest campaign
that probably adresses some poor indians that
have neither 5G nor Macs:

1:38:01 The Kyiv keynote disaster
https://www.youtube.com/watch?v=U8goS6B3BbI

Woa! Real time download of Scala, Closure,
etc.. Whats the magic behind that? Some SWI
point of sale, downloading it via its

keyboard and some telephathy module ?

Bye

Mild Shock schrieb:

Hi,

Ride the snake
He's old and his skin is cold
The west is the best
The west is the best
Get here and we'll do the rest
The blue bus is calling us
The blue bus is calling us
Driver, where you taking us?

Apocalypse Now intro: The Doors, The End {1979}
https://www.youtube.com/watch?v=CIrvSJwwJUE

Bye

Hi,

Again I posted everything here:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

The repo says, same time when I posted
the link first time:

This repository was archived by the
owner on Jul 9, 2026. It is now read-only.

Now a USENET user, who had already entitled
himself for a couple of irrational accusations

towards my side, is asking this question:

Chris M. Thomasson schrieb, Jul 24, 2026

Show an outline of what you
need you compute shader to do?

Bravo, thats a delay of a wooping 15 days.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

There is a typo here:

Subject: I wrote Hack VM for -C-WAM from scratch
[4 Months total JavaScript, Python and Java]

It should say:

Subject: I wrote Hack VM for -C-WAM from scratch
[4 Weeks total JavaScript, Python and Java]

Sorry!

Bye

Mild Shock schrieb:

Hi,

The op-codes are all uniform, have the
same sub fields. Already Z-80 CPU differs here.
Another difference to a Z-80 CPU is that

their instruction stream was 8-bit, instructions
can 1, 2, 3 or 4 byte long. On the other
hand in my Hack VM all instructions are

one 32-bit chunk. The porting of a first
prototype that I already had, to WebGPU / WGSL
only took like 1-2 hours. The execution

of Hack VM is very simple, version 1.0,
for a single shader:

fn run() {
-a-a-a var pc : i32 = 0;
-a-a-a var accu : i32 = 0;
-a-a-a while (pc < i32(arrayLength(&code))) {
-a-a-a-a-a-a-a var instr : i32 = code[pc];
-a-a-a-a-a-a-a pc += 1;
-a-a-a-a-a-a-a var value : i32 = run_get(instr);
-a-a-a-a-a-a-a accu = run_fun(instr, accu, value);
-a-a-a-a-a-a-a run_set(instr, accu);
-a-a-a-a-a-a-a pc += run_jump(instr, accu);
-a-a-a }
}

https://github.com/Jean-Luc-Picard-2021/gigabudget/blob/b8946e891be774c40522267ab17062d32b023e7a/course/example63/boot.mjs#L176-L187

I first though this will be perfect for
SIMD. Until I learnt that modern GPUs have
anyway MIMD. Hell Yeah, thats much better!

Bye

Mild Shock schrieb:

Hi,

Just downloading some other person's code

I didn't do that, I wrote Hack VM for pi-WAM
from scratch, over the last 4 weeks. I came
back from holidays on end of June 2026, and now

we have end of July 2026. But its only possible
because the instruction set is very smal, like
ca. 8 functions and ca. 8 modes and ca. 8 conditions,

so its ca. 8 x 8 x 8 = 512 opcodes, each has an
A parameter and a D parameter simultaneously.
It has currently the following CPU backends:

-a-a- Now supports interleaved synchronous emulation.
-a-a- Now supports warp parallelism via Java platform threads.
-a-a- Now supports warp parallelism via Python system threads.
-a-a- Now supports warp parallelism via JavaScript worker threads.
-a-a- Note: For Python free threads are not yet fully tested.
-a-a- Note: For JavaScript web workers are not yet fully tested.

https://www.dogelog.ch/typtab/doclet/book/14_install/05_notes22/110_224.html


But frankly I came to encounter Hack not from
the usual university curriculum web resources,
but indirectly through a post about a Prolog

emulation of Hack, using constrained horn clauses (CHC):

Verifying Nand2Tetris Assembly
https://www.philipzucker.com/nand2tetris-chc/

The binary encoding is currently that the functions,
modes and conditions eat up a nibble (4-bit), in
total 12-bit, which I use then 10-bit for A parameter

and 10-bit for D parameter. I used AI freemium, Codex
by ChatGPT from within IntelliJ to do some fragment
code translations automatically from Java to JavaScript

or from JavaScript to Python.

Have Fun!

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

There is Prolog compiler which spits out Hack.
From there on your are free to develop
and/or use any Hack realization that goes

from abstract to concrete. You could
replace the CPU backends that realize
a Hack VM by MIPS. Shouldn't be difficult.

Basically I refused to think in Huffman
Coding (*) while designing Hack VM. On the
other hand the MIPS architecture looks

like a big Huffman mess. Already its
initial design has 3 instructions types:

Type format (bits)
R opcode(6) rs(5) rt(5) rd(5) shamt(5) funct(6)
I opcode(6) rs(5) rt(5) imme(16)
J opcode(6) addr(26)

While my Hack has only 1 instruction
type, when binary encoded for Hack VM,
the currently used design looks as follows:

Type format (bits)
AD opcode(4) mode(4) cond(4) imme(10) addr(10)

But since its an abstract machine, nothing
prevents you from translating Hack code
into MIPS before executing it.

In has far you have to distinguish Hack,
which is specified in Prolog. And Hack VM
which is a virtual machine, with the above

instruction packing. And which has currently
a JavaScript runtime, a Python runtime
and a Java runtime.

Bye

(*)
https://en.wikipedia.org/wiki/Huffman_coding

Mild Shock schrieb:

Hi,

Just downloading some other person's code

I didn't do that, I wrote Hack VM for pi-WAM
from scratch, over the last 4 weeks. I came
back from holidays on end of June 2026, and now

we have end of July 2026. But its only possible
because the instruction set is very smal, like
ca. 8 functions and ca. 8 modes and ca. 8 conditions,

so its ca. 8 x 8 x 8 = 512 opcodes, each has an
A parameter and a D parameter simultaneously.
It has currently the following CPU backends:

-a- Now supports interleaved synchronous emulation.
-a- Now supports warp parallelism via Java platform threads.
-a- Now supports warp parallelism via Python system threads.
-a- Now supports warp parallelism via JavaScript worker threads.
-a- Note: For Python free threads are not yet fully tested.
-a- Note: For JavaScript web workers are not yet fully tested.

https://www.dogelog.ch/typtab/doclet/book/14_install/05_notes22/110_224.html

But frankly I came to encounter Hack not from
the usual university curriculum web resources,
but indirectly through a post about a Prolog

emulation of Hack, using constrained horn clauses (CHC):

Verifying Nand2Tetris Assembly
https://www.philipzucker.com/nand2tetris-chc/

The binary encoding is currently that the functions,
modes and conditions eat up a nibble (4-bit), in
total 12-bit, which I use then 10-bit for A parameter

and 10-bit for D parameter. I used AI freemium, Codex
by ChatGPT from within IntelliJ to do some fragment
code translations automatically from Java to JavaScript

or from JavaScript to Python.

Have Fun!

Bye

Mild Shock schrieb:

Hi,

Who exactly is the thief? Does this person
have stats in the Rogue class in dungeons
and dragons?

The conspiracy theory of a stealing of Torso VDBE,
by Rossy Boy, is probably a result of complete
ignorance of the Hack ecosystem.

Hack is a very popular computer science project,
with a couple of subprojects in hardware and
software. It goes also by the name Nand to Tetris,

and is programming language agnositic. You can do
Hack experiments in any programming language, be
it BASIC, ADA or Rust. Nobody cares.

The gist are projects like here, first to
educate yourself about Hack:

https://www.nand2tetris.org/course

And then to use Hack in different contexts:

https://www.nand2tetris.org/copy-of-talks

For didactic purposes, I used Hack for my WebGPU
experiment. I didn't even take a look at Torso
VDBE, why should I? Hack is nicely documented,

has even a book, and fusing the two 16-bit
instruction types A and D, into a single 32-bit
instruction stream, is nowhere patented.

Bye

Mild Shock schrieb:

Hi,

I don't use Rust, you are crazy. First of
all the parallel simulator is 100% written
in Prolog, should also run in ISO Prolog,

enhanced by a library(lists). Second I only
mentioned that WebGPU / WGSL, the language
there has a Rust inspired language.

Its not Rust. Whats wrong with you? Why do
you adress your weariness of life to me.
I am neither thief, nor can I help you

with your frustration, and histeric outbursts.
Maybe just be a man and jump off a bridge, idiot.
Or tame your frustration, usenet is not for

you alone, your stupid asshole.

Bye

Ross Finlayson schrieb:

https://www.theregister.com/databases/2026/07/29/after-rewriting-sqlite-in-rust-turso-turns-its-sights-on-postgres/5279835

I don't much care about Rust.

.. gibberish ..

Thief.

Mild Shock schrieb:

Hi,

Rossy Boys tears could cool a data center,
he thinks there exists no literature about
serial algorithms of parallel stuff, and

he also thinks normal forms lead to optimizing
something. LoL, what a utter bullshit. I did
alreay a serial implementation of a parallel

simulation of my pi-WAM. Just lookup the literature
about pi-calulus. I published it a few days ago,
its part of 2.2.4 released already:

Parallel -C-WAM: An Interleaved Synchronous Emulator
https://medium.com/2989/0196089e143a

Whats your point, Rossy Boy? Except you post pretend
nonsense not knowing what you are doing?

Bye

Ross Finlayson schrieb:

No, troll, these are serial algorithms their optimized forms.

Normal sorts of forms, ....


Yeah, everybody already figured out "interpreters" and
"programs" and "spawning".

Go spawn yourself.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of

less than

a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one microsecond. >>>>>

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, that >>>>> -a> the indicators of the above as "positive presence" then is to make >>>>> -a> for that the adjustments to the offsets and extents and the shifts >>>>> -a> is according to those, otherwise no-ops. Then the idea is that a >>>>>

As independent logical thread state, that automatically MIMD follows? >>>>>
Whats the problem to solve then?

Bye

Mild Shock schrieb:

Hi,

Rossy Boy is neither Einstein nor Zweistein.
He is not Einstein since Einstein is already dead:

Albert Einstein (1879 - 1955)
https://de.wikipedia.org/wiki/Albert_Einstein

He is also not Zweistein, since he doesn't
understand concepts such as:

- NVIDIA Volta ff. architecture

Also his hands are small, and his breath stinks,
and he lives in the basement of his mother.

Bye

Mild Shock schrieb:

Hi,

Slowly I start understanding numbnuts like
Rossy Boy who don't understand tech, although
they are from UK and not from a 3rd world

country, and also I start understanding morons
like Micro Penis, who are behind a curtain,
and cannot access a lot of tech.

The same holds for SWI Prologs newest campaign
that probably adresses some poor indians that
have neither 5G nor Macs:

1:38:01 The Kyiv keynote disaster
https://www.youtube.com/watch?v=U8goS6B3BbI

Woa! Real time download of Scala, Closure,
etc.. Whats the magic behind that? Some SWI
point of sale, downloading it via its

keyboard and some telephathy module ?

Bye

Mild Shock schrieb:

Hi,

Ride the snake
He's old and his skin is cold
The west is the best
The west is the best
Get here and we'll do the rest
The blue bus is calling us
The blue bus is calling us
Driver, where you taking us?

Apocalypse Now intro: The Doors, The End {1979}
https://www.youtube.com/watch?v=CIrvSJwwJUE

Bye

Hi,

Again I posted everything here:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

The repo says, same time when I posted
the link first time:

This repository was archived by the
owner on Jul 9, 2026. It is now read-only.

Now a USENET user, who had already entitled
himself for a couple of irrational accusations

towards my side, is asking this question:

Chris M. Thomasson schrieb, Jul 24, 2026

Show an outline of what you
need you compute shader to do?

Bravo, thats a delay of a wooping 15 days.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Another choice for naming Hack, would be
to call it an intermediate format. But
this is typically used here:

The intermediate representation, or
IR for short, is an in-memory data
structure that represents executable code. https://www.llvmpy.org/llvmpy-doc/dev/doc/llvm_concepts.html#ssa-form-and-phi-nodes

So I still like the term abstract machine,
as already used in the past by David H. D. Warren
for the famous, and in my opinion infamous:

Warren Abstract Machine 1983 https://en.wikipedia.org/wiki/Warren_Abstract_Machine

Maybe you can take the term abstract machine
as a hint that it is more lower level, and
more imperative. Not something highlevel, that

is easily malleable. But abstract also captures
the notion that there is still a level further
down, making it concrete. And you find

many Prolog systems that did just that, they
compile WAM into a further instruction stream,
like x86 or whatever, for binary compiled code,

that is not interpreted WAM.

Bye

Mild Shock schrieb:

Hi,

There is Prolog compiler which spits out Hack.
From there on your are free to develop
and/or use any Hack realization that goes

from abstract to concrete. You could
replace the CPU backends that realize
a Hack VM by MIPS. Shouldn't be difficult.

Basically I refused to think in Huffman
Coding (*) while designing Hack VM. On the
other hand the MIPS architecture looks

like a big Huffman mess. Already its
initial design has 3 instructions types:

Type format (bits)
R opcode(6) rs(5) rt(5) rd(5) shamt(5) funct(6)
I opcode(6) rs(5) rt(5) imme(16)
J opcode(6) addr(26)

While my Hack has only 1 instruction
type, when binary encoded for Hack VM,
the currently used design looks as follows:

Type format (bits)
AD opcode(4) mode(4) cond(4) imme(10) addr(10)

But since its an abstract machine, nothing
prevents you from translating Hack code
into MIPS before executing it.

In has far you have to distinguish Hack,
which is specified in Prolog. And Hack VM
which is a virtual machine, with the above

instruction packing. And which has currently
a JavaScript runtime, a Python runtime
and a Java runtime.

Bye

(*)
https://en.wikipedia.org/wiki/Huffman_coding

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Another choice for naming Hack, would be
to call it an intermediate format. But
this is typically used here:

The intermediate representation, or
IR for short, is an in-memory data
structure that represents executable code. https://www.llvmpy.org/llvmpy-doc/dev/doc/llvm_concepts.html#ssa-form-and-phi-nodes

So I still like the term abstract machine,
as already used in the past by David H. D. Warren
for the famous, and in my opinion infamous:

Warren Abstract Machine 1983 https://en.wikipedia.org/wiki/Warren_Abstract_Machine

Maybe you can take the term abstract machine
as a hint that it is more lower level, and
more imperative. Not something highlevel, that

is easily malleable. But abstract also captures
the notion that there is still a level further
down, making it concrete. And you find

many Prolog systems that did just that, they
compile WAM into a further instruction stream,
like x86 or whatever, for binary compiled code,

that is not interpreted WAM.

Bye

Mild Shock schrieb:

Hi,

There is Prolog compiler which spits out Hack.
From there on your are free to develop
and/or use any Hack realization that goes

from abstract to concrete. You could
replace the CPU backends that realize
a Hack VM by MIPS. Shouldn't be difficult.

Basically I refused to think in Huffman
Coding (*) while designing Hack VM. On the
other hand the MIPS architecture looks

like a big Huffman mess. Already its
initial design has 3 instructions types:

Type format (bits)
R opcode(6) rs(5) rt(5) rd(5) shamt(5) funct(6)
I opcode(6) rs(5) rt(5) imme(16)
J opcode(6) addr(26)

While my Hack has only 1 instruction
type, when binary encoded for Hack VM,
the currently used design looks as follows:

Type format (bits)
AD opcode(4) mode(4) cond(4) imme(10) addr(10)

But since its an abstract machine, nothing
prevents you from translating Hack code
into MIPS before executing it.

In has far you have to distinguish Hack,
which is specified in Prolog. And Hack VM
which is a virtual machine, with the above

instruction packing. And which has currently
a JavaScript runtime, a Python runtime
and a Java runtime.

Bye

(*)
https://en.wikipedia.org/wiki/Huffman_coding

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

On 07/30/2026 01:32 PM, Mild Shock wrote:

Hi,

There is Prolog compiler which spits out Hack.
From there on your are free to develop
and/or use any Hack realization that goes

from abstract to concrete. You could
replace the CPU backends that realize
a Hack VM by MIPS. Shouldn't be difficult.

Basically I refused to think in Huffman
Coding (*) while designing Hack VM. On the
other hand the MIPS architecture looks

like a big Huffman mess. Already its
initial design has 3 instructions types:

Type format (bits)
R opcode(6) rs(5) rt(5) rd(5) shamt(5) funct(6)
I opcode(6) rs(5) rt(5) imme(16)
J opcode(6) addr(26)

While my Hack has only 1 instruction
type, when binary encoded for Hack VM,
the currently used design looks as follows:

Type format (bits)
AD opcode(4) mode(4) cond(4) imme(10) addr(10)

But since its an abstract machine, nothing
prevents you from translating Hack code
into MIPS before executing it.

In has far you have to distinguish Hack,
which is specified in Prolog. And Hack VM
which is a virtual machine, with the above

instruction packing. And which has currently
a JavaScript runtime, a Python runtime
and a Java runtime.

Bye

(*)
https://en.wikipedia.org/wiki/Huffman_coding

Mild Shock schrieb:

Hi,

Just downloading some other person's code

I didn't do that, I wrote Hack VM for pi-WAM
from scratch, over the last 4 weeks. I came
back from holidays on end of June 2026, and now

we have end of July 2026. But its only possible
because the instruction set is very smal, like
ca. 8 functions and ca. 8 modes and ca. 8 conditions,

so its ca. 8 x 8 x 8 = 512 opcodes, each has an
A parameter and a D parameter simultaneously.
It has currently the following CPU backends:

- Now supports interleaved synchronous emulation.
- Now supports warp parallelism via Java platform threads.
- Now supports warp parallelism via Python system threads.
- Now supports warp parallelism via JavaScript worker threads.
- Note: For Python free threads are not yet fully tested.
- Note: For JavaScript web workers are not yet fully tested.

https://www.dogelog.ch/typtab/doclet/book/14_install/05_notes22/110_224.html >>

But frankly I came to encounter Hack not from
the usual university curriculum web resources,
but indirectly through a post about a Prolog

emulation of Hack, using constrained horn clauses (CHC):

Verifying Nand2Tetris Assembly
https://www.philipzucker.com/nand2tetris-chc/

The binary encoding is currently that the functions,
modes and conditions eat up a nibble (4-bit), in
total 12-bit, which I use then 10-bit for A parameter

and 10-bit for D parameter. I used AI freemium, Codex
by ChatGPT from within IntelliJ to do some fragment
code translations automatically from Java to JavaScript

or from JavaScript to Python.

Have Fun!

Bye

Mild Shock schrieb:

Hi,

Who exactly is the thief? Does this person
have stats in the Rogue class in dungeons
and dragons?

The conspiracy theory of a stealing of Torso VDBE,
by Rossy Boy, is probably a result of complete
ignorance of the Hack ecosystem.

Hack is a very popular computer science project,
with a couple of subprojects in hardware and
software. It goes also by the name Nand to Tetris,

and is programming language agnositic. You can do
Hack experiments in any programming language, be
it BASIC, ADA or Rust. Nobody cares.

The gist are projects like here, first to
educate yourself about Hack:

https://www.nand2tetris.org/course

And then to use Hack in different contexts:

https://www.nand2tetris.org/copy-of-talks

For didactic purposes, I used Hack for my WebGPU
experiment. I didn't even take a look at Torso
VDBE, why should I? Hack is nicely documented,

has even a book, and fusing the two 16-bit
instruction types A and D, into a single 32-bit
instruction stream, is nowhere patented.

Bye

Mild Shock schrieb:

Hi,

I don't use Rust, you are crazy. First of
all the parallel simulator is 100% written
in Prolog, should also run in ISO Prolog,

enhanced by a library(lists). Second I only
mentioned that WebGPU / WGSL, the language
there has a Rust inspired language.

Its not Rust. Whats wrong with you? Why do
you adress your weariness of life to me.
I am neither thief, nor can I help you

with your frustration, and histeric outbursts.
Maybe just be a man and jump off a bridge, idiot.
Or tame your frustration, usenet is not for

you alone, your stupid asshole.

Bye

Ross Finlayson schrieb:

https://www.theregister.com/databases/2026/07/29/after-rewriting-sqlite-in-rust-turso-turns-its-sights-on-postgres/5279835

I don't much care about Rust.

.. gibberish ..

Thief.

Mild Shock schrieb:

Hi,

Rossy Boys tears could cool a data center,
he thinks there exists no literature about
serial algorithms of parallel stuff, and

he also thinks normal forms lead to optimizing
something. LoL, what a utter bullshit. I did
alreay a serial implementation of a parallel

simulation of my pi-WAM. Just lookup the literature
about pi-calulus. I published it a few days ago,
its part of 2.2.4 released already:

Parallel -C-WAM: An Interleaved Synchronous Emulator
https://medium.com/2989/0196089e143a

Whats your point, Rossy Boy? Except you post pretend
nonsense not knowing what you are doing?

Bye

Ross Finlayson schrieb:

No, troll, these are serial algorithms their optimized forms.

Normal sorts of forms, ....


Yeah, everybody already figured out "interpreters" and
"programs" and "spawning".

Go spawn yourself.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of

less than

a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one

microsecond.

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents,

that

the indicators of the above as "positive presence" then is to make >>>>>> > for that the adjustments to the offsets and extents and the shifts >>>>>> > is according to those, otherwise no-ops. Then the idea is that a >>>>>>

As independent logical thread state, that automatically MIMD follows? >>>>>>
Whats the problem to solve then?

Bye

Mild Shock schrieb:

Hi,

Rossy Boy is neither Einstein nor Zweistein.
He is not Einstein since Einstein is already dead:

Albert Einstein (1879 - 1955)
https://de.wikipedia.org/wiki/Albert_Einstein

He is also not Zweistein, since he doesn't
understand concepts such as:

- NVIDIA Volta ff. architecture

Also his hands are small, and his breath stinks,
and he lives in the basement of his mother.

Bye

Mild Shock schrieb:

Hi,

Slowly I start understanding numbnuts like
Rossy Boy who don't understand tech, although
they are from UK and not from a 3rd world

country, and also I start understanding morons
like Micro Penis, who are behind a curtain,
and cannot access a lot of tech.

The same holds for SWI Prologs newest campaign
that probably adresses some poor indians that
have neither 5G nor Macs:

1:38:01 The Kyiv keynote disaster
https://www.youtube.com/watch?v=U8goS6B3BbI

Woa! Real time download of Scala, Closure,
etc.. Whats the magic behind that? Some SWI
point of sale, downloading it via its

keyboard and some telephathy module ?

Bye

Mild Shock schrieb:

Hi,

Ride the snake
He's old and his skin is cold
The west is the best
The west is the best
Get here and we'll do the rest
The blue bus is calling us
The blue bus is calling us
Driver, where you taking us?

Apocalypse Now intro: The Doors, The End {1979}
https://www.youtube.com/watch?v=CIrvSJwwJUE

Bye

Hi,

Again I posted everything here:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

The repo says, same time when I posted
the link first time:

This repository was archived by the
owner on Jul 9, 2026. It is now read-only.

Now a USENET user, who had already entitled
himself for a couple of irrational accusations

towards my side, is asking this question:

Chris M. Thomasson schrieb, Jul 24, 2026

Show an outline of what you
need you compute shader to do?

Bravo, thats a delay of a wooping 15 days.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

Huffman was a bright lad who it's said came up with Huffman
coding to get past taking a final arduous exam.

One of my grandfather's name is Huffman, he graduated high-school,
somewhere's a picture of him with his high-school graduating
class, in Cleveland, Ohio, more than a hundred years ago.

Three people graduated high-school that year.

Huffman codes are a fantastic thing and ubiquitous, though,
like any other algorithm & data structure, they depend on
the distribution their best/average/worst cases, since everybody
here knows the general outline of asymptotics and Big O, Little o,
and Theta, about time and space and complexity.


Then, "adaptive algorithms" and "adapative data structures",
start with something like "compression" in usual accounts,
which usually is adaptive in the sense of compressing what's
compressible in a window, and making a histogram of the words
in the alphabet interpreted as a sequence of letters, and making
an optimal sort of Huffman-coding for entropy-coding for that,
since Huffman-coding is naturally optimal for a particular distribution,
that like power laws can be found in anything,
that one starts with making histograms after combinatorial enumeration,
and thus resulting an encoding into a language of bit-sequences
with the prefix-property that un-ambiguously makes for compressing
the compressible data.

Then the "adaptive" part of that is
periodically throwing that away and starting another.
One might aver that a more contextually-advised account
studies the entire corpus, for things like CCITT G4, JPEG,
JBIG, Deflate, and any account of entropy-coding or compression,
or Morse code, with of course both Huffman-coding and arithmetic-coding,
since while Huffman-coding is obvious to everybody,
for a while some people thought arithmetic coding had patents,
which now are gone away, leaving all the above mentioned
and MPEG-4 also the "un-encumbered".

The notion of "summary statistics" and "order statistics" though,
in concrete mathematics of course is simple and clear.


Here's an article I read the other day about Huffman-coding,
I found it very insightful and quite enjoyable.

https://fgiesen.wordpress.com/2026/06/21/pivco-huffman-merge-operations/




Shut Up


--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

On 07/31/2026 09:52 AM, Ross Finlayson wrote:

On 07/30/2026 01:32 PM, Mild Shock wrote:

Hi,

There is Prolog compiler which spits out Hack.
From there on your are free to develop
and/or use any Hack realization that goes

from abstract to concrete. You could
replace the CPU backends that realize
a Hack VM by MIPS. Shouldn't be difficult.

Basically I refused to think in Huffman
Coding (*) while designing Hack VM. On the
other hand the MIPS architecture looks

like a big Huffman mess. Already its
initial design has 3 instructions types:

Type format (bits)
R opcode(6) rs(5) rt(5) rd(5) shamt(5) funct(6)
I opcode(6) rs(5) rt(5) imme(16)
J opcode(6) addr(26)

While my Hack has only 1 instruction
type, when binary encoded for Hack VM,
the currently used design looks as follows:

Type format (bits)
AD opcode(4) mode(4) cond(4) imme(10) addr(10)

But since its an abstract machine, nothing
prevents you from translating Hack code
into MIPS before executing it.

In has far you have to distinguish Hack,
which is specified in Prolog. And Hack VM
which is a virtual machine, with the above

instruction packing. And which has currently
a JavaScript runtime, a Python runtime
and a Java runtime.

Bye

(*)
https://en.wikipedia.org/wiki/Huffman_coding

Mild Shock schrieb:

Hi,

Just downloading some other person's code

I didn't do that, I wrote Hack VM for pi-WAM
from scratch, over the last 4 weeks. I came
back from holidays on end of June 2026, and now

we have end of July 2026. But its only possible
because the instruction set is very smal, like
ca. 8 functions and ca. 8 modes and ca. 8 conditions,

so its ca. 8 x 8 x 8 = 512 opcodes, each has an
A parameter and a D parameter simultaneously.
It has currently the following CPU backends:

- Now supports interleaved synchronous emulation.
- Now supports warp parallelism via Java platform threads.
- Now supports warp parallelism via Python system threads.
- Now supports warp parallelism via JavaScript worker threads.
- Note: For Python free threads are not yet fully tested.
- Note: For JavaScript web workers are not yet fully tested.

https://www.dogelog.ch/typtab/doclet/book/14_install/05_notes22/110_224.html



But frankly I came to encounter Hack not from
the usual university curriculum web resources,
but indirectly through a post about a Prolog

emulation of Hack, using constrained horn clauses (CHC):

Verifying Nand2Tetris Assembly
https://www.philipzucker.com/nand2tetris-chc/

The binary encoding is currently that the functions,
modes and conditions eat up a nibble (4-bit), in
total 12-bit, which I use then 10-bit for A parameter

and 10-bit for D parameter. I used AI freemium, Codex
by ChatGPT from within IntelliJ to do some fragment
code translations automatically from Java to JavaScript

or from JavaScript to Python.

Have Fun!

Bye

Mild Shock schrieb:

Hi,

Who exactly is the thief? Does this person
have stats in the Rogue class in dungeons
and dragons?

The conspiracy theory of a stealing of Torso VDBE,
by Rossy Boy, is probably a result of complete
ignorance of the Hack ecosystem.

Hack is a very popular computer science project,
with a couple of subprojects in hardware and
software. It goes also by the name Nand to Tetris,

and is programming language agnositic. You can do
Hack experiments in any programming language, be
it BASIC, ADA or Rust. Nobody cares.

The gist are projects like here, first to
educate yourself about Hack:

https://www.nand2tetris.org/course

And then to use Hack in different contexts:

https://www.nand2tetris.org/copy-of-talks

For didactic purposes, I used Hack for my WebGPU
experiment. I didn't even take a look at Torso
VDBE, why should I? Hack is nicely documented,

has even a book, and fusing the two 16-bit
instruction types A and D, into a single 32-bit
instruction stream, is nowhere patented.

Bye

Mild Shock schrieb:

Hi,

I don't use Rust, you are crazy. First of
all the parallel simulator is 100% written
in Prolog, should also run in ISO Prolog,

enhanced by a library(lists). Second I only
mentioned that WebGPU / WGSL, the language
there has a Rust inspired language.

Its not Rust. Whats wrong with you? Why do
you adress your weariness of life to me.
I am neither thief, nor can I help you

with your frustration, and histeric outbursts.
Maybe just be a man and jump off a bridge, idiot.
Or tame your frustration, usenet is not for

you alone, your stupid asshole.

Bye

Ross Finlayson schrieb:

https://www.theregister.com/databases/2026/07/29/after-rewriting-sqlite-in-rust-turso-turns-its-sights-on-postgres/5279835

I don't much care about Rust.

.. gibberish ..

Thief.

Mild Shock schrieb:

Hi,

Rossy Boys tears could cool a data center,
he thinks there exists no literature about
serial algorithms of parallel stuff, and

he also thinks normal forms lead to optimizing
something. LoL, what a utter bullshit. I did
alreay a serial implementation of a parallel

simulation of my pi-WAM. Just lookup the literature
about pi-calulus. I published it a few days ago,
its part of 2.2.4 released already:

Parallel -C-WAM: An Interleaved Synchronous Emulator
https://medium.com/2989/0196089e143a

Whats your point, Rossy Boy? Except you post pretend
nonsense not knowing what you are doing?

Bye

Ross Finlayson schrieb:

No, troll, these are serial algorithms their optimized forms.

Normal sorts of forms, ....


Yeah, everybody already figured out "interpreters" and
"programs" and "spawning".

Go spawn yourself.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in >>>>>>> > one "run", i.e. a stall-less, branch-less, call-less list of

less than

a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding >>>>>>> > "matchings" of expressions, that runs in less than one

microsecond.

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, >>>>>>> that
the indicators of the above as "positive presence" then is to >>>>>>> make
for that the adjustments to the offsets and extents and the

shifts

is according to those, otherwise no-ops. Then the idea is that a >>>>>>>

As independent logical thread state, that automatically MIMD
follows?

Whats the problem to solve then?

Bye

Mild Shock schrieb:

Hi,

Rossy Boy is neither Einstein nor Zweistein.
He is not Einstein since Einstein is already dead:

Albert Einstein (1879 - 1955)
https://de.wikipedia.org/wiki/Albert_Einstein

He is also not Zweistein, since he doesn't
understand concepts such as:

- NVIDIA Volta ff. architecture

Also his hands are small, and his breath stinks,
and he lives in the basement of his mother.

Bye

Mild Shock schrieb:

Hi,

Slowly I start understanding numbnuts like
Rossy Boy who don't understand tech, although
they are from UK and not from a 3rd world

country, and also I start understanding morons
like Micro Penis, who are behind a curtain,
and cannot access a lot of tech.

The same holds for SWI Prologs newest campaign
that probably adresses some poor indians that
have neither 5G nor Macs:

1:38:01 The Kyiv keynote disaster
https://www.youtube.com/watch?v=U8goS6B3BbI

Woa! Real time download of Scala, Closure,
etc.. Whats the magic behind that? Some SWI
point of sale, downloading it via its

keyboard and some telephathy module ?

Bye

Mild Shock schrieb:

Hi,

Ride the snake
He's old and his skin is cold
The west is the best
The west is the best
Get here and we'll do the rest
The blue bus is calling us
The blue bus is calling us
Driver, where you taking us?

Apocalypse Now intro: The Doors, The End {1979}
https://www.youtube.com/watch?v=CIrvSJwwJUE

Bye

Hi,

Again I posted everything here:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

The repo says, same time when I posted
the link first time:

This repository was archived by the
owner on Jul 9, 2026. It is now read-only.

Now a USENET user, who had already entitled
himself for a couple of irrational accusations

towards my side, is asking this question:

Chris M. Thomasson schrieb, Jul 24, 2026

Show an outline of what you
need you compute shader to do?

Bravo, thats a delay of a wooping 15 days.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture >>>>>>>>>>> with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

Huffman was a bright lad who it's said came up with Huffman
coding to get past taking a final arduous exam.

One of my grandfather's name is Huffman, he graduated high-school, somewhere's a picture of him with his high-school graduating
class, in Cleveland, Ohio, more than a hundred years ago.

Three people graduated high-school that year.

Huffman codes are a fantastic thing and ubiquitous, though,
like any other algorithm & data structure, they depend on
the distribution their best/average/worst cases, since everybody
here knows the general outline of asymptotics and Big O, Little o,
and Theta, about time and space and complexity.

Then, "adaptive algorithms" and "adapative data structures",
start with something like "compression" in usual accounts,
which usually is adaptive in the sense of compressing what's
compressible in a window, and making a histogram of the words
in the alphabet interpreted as a sequence of letters, and making
an optimal sort of Huffman-coding for entropy-coding for that,
since Huffman-coding is naturally optimal for a particular distribution,
that like power laws can be found in anything,
that one starts with making histograms after combinatorial enumeration,
and thus resulting an encoding into a language of bit-sequences
with the prefix-property that un-ambiguously makes for compressing
the compressible data.

Then the "adaptive" part of that is
periodically throwing that away and starting another.
One might aver that a more contextually-advised account
studies the entire corpus, for things like CCITT G4, JPEG,
JBIG, Deflate, and any account of entropy-coding or compression,
or Morse code, with of course both Huffman-coding and arithmetic-coding, since while Huffman-coding is obvious to everybody,
for a while some people thought arithmetic coding had patents,
which now are gone away, leaving all the above mentioned
and MPEG-4 also the "un-encumbered".

The notion of "summary statistics" and "order statistics" though,
in concrete mathematics of course is simple and clear.

Here's an article I read the other day about Huffman-coding,
I found it very insightful and quite enjoyable.

https://fgiesen.wordpress.com/2026/06/21/pivco-huffman-merge-operations/

Shut Up

"It easily scales both up and down
with the capabilities of the target machine."

- https://fgiesen.wordpress.com/2026/06/21/pivco-huffman-merge-operations/


--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

On 07/31/2026 10:00 AM, Ross Finlayson wrote:

On 07/31/2026 09:52 AM, Ross Finlayson wrote:

On 07/30/2026 01:32 PM, Mild Shock wrote:

Hi,

There is Prolog compiler which spits out Hack.
From there on your are free to develop
and/or use any Hack realization that goes

from abstract to concrete. You could
replace the CPU backends that realize
a Hack VM by MIPS. Shouldn't be difficult.

Basically I refused to think in Huffman
Coding (*) while designing Hack VM. On the
other hand the MIPS architecture looks

like a big Huffman mess. Already its
initial design has 3 instructions types:

Type format (bits)
R opcode(6) rs(5) rt(5) rd(5) shamt(5) funct(6)
I opcode(6) rs(5) rt(5) imme(16)
J opcode(6) addr(26)

While my Hack has only 1 instruction
type, when binary encoded for Hack VM,
the currently used design looks as follows:

Type format (bits)
AD opcode(4) mode(4) cond(4) imme(10) addr(10)

But since its an abstract machine, nothing
prevents you from translating Hack code
into MIPS before executing it.

In has far you have to distinguish Hack,
which is specified in Prolog. And Hack VM
which is a virtual machine, with the above

instruction packing. And which has currently
a JavaScript runtime, a Python runtime
and a Java runtime.

Bye

(*)
https://en.wikipedia.org/wiki/Huffman_coding

Mild Shock schrieb:

Hi,

Just downloading some other person's code

I didn't do that, I wrote Hack VM for pi-WAM
from scratch, over the last 4 weeks. I came
back from holidays on end of June 2026, and now

we have end of July 2026. But its only possible
because the instruction set is very smal, like
ca. 8 functions and ca. 8 modes and ca. 8 conditions,

so its ca. 8 x 8 x 8 = 512 opcodes, each has an
A parameter and a D parameter simultaneously.
It has currently the following CPU backends:

- Now supports interleaved synchronous emulation.
- Now supports warp parallelism via Java platform threads.
- Now supports warp parallelism via Python system threads.
- Now supports warp parallelism via JavaScript worker threads.
- Note: For Python free threads are not yet fully tested.
- Note: For JavaScript web workers are not yet fully tested.

https://www.dogelog.ch/typtab/doclet/book/14_install/05_notes22/110_224.html




But frankly I came to encounter Hack not from
the usual university curriculum web resources,
but indirectly through a post about a Prolog

emulation of Hack, using constrained horn clauses (CHC):

Verifying Nand2Tetris Assembly
https://www.philipzucker.com/nand2tetris-chc/

The binary encoding is currently that the functions,
modes and conditions eat up a nibble (4-bit), in
total 12-bit, which I use then 10-bit for A parameter

and 10-bit for D parameter. I used AI freemium, Codex
by ChatGPT from within IntelliJ to do some fragment
code translations automatically from Java to JavaScript

or from JavaScript to Python.

Have Fun!

Bye

Mild Shock schrieb:

Hi,

Who exactly is the thief? Does this person
have stats in the Rogue class in dungeons
and dragons?

The conspiracy theory of a stealing of Torso VDBE,
by Rossy Boy, is probably a result of complete
ignorance of the Hack ecosystem.

Hack is a very popular computer science project,
with a couple of subprojects in hardware and
software. It goes also by the name Nand to Tetris,

and is programming language agnositic. You can do
Hack experiments in any programming language, be
it BASIC, ADA or Rust. Nobody cares.

The gist are projects like here, first to
educate yourself about Hack:

https://www.nand2tetris.org/course

And then to use Hack in different contexts:

https://www.nand2tetris.org/copy-of-talks

For didactic purposes, I used Hack for my WebGPU
experiment. I didn't even take a look at Torso
VDBE, why should I? Hack is nicely documented,

has even a book, and fusing the two 16-bit
instruction types A and D, into a single 32-bit
instruction stream, is nowhere patented.

Bye

Mild Shock schrieb:

Hi,

I don't use Rust, you are crazy. First of
all the parallel simulator is 100% written
in Prolog, should also run in ISO Prolog,

enhanced by a library(lists). Second I only
mentioned that WebGPU / WGSL, the language
there has a Rust inspired language.

Its not Rust. Whats wrong with you? Why do
you adress your weariness of life to me.
I am neither thief, nor can I help you

with your frustration, and histeric outbursts.
Maybe just be a man and jump off a bridge, idiot.
Or tame your frustration, usenet is not for

you alone, your stupid asshole.

Bye

Ross Finlayson schrieb:

https://www.theregister.com/databases/2026/07/29/after-rewriting-sqlite-in-rust-turso-turns-its-sights-on-postgres/5279835

I don't much care about Rust.

.. gibberish ..

Thief.

Mild Shock schrieb:

Hi,

Rossy Boys tears could cool a data center,
he thinks there exists no literature about
serial algorithms of parallel stuff, and

he also thinks normal forms lead to optimizing
something. LoL, what a utter bullshit. I did
alreay a serial implementation of a parallel

simulation of my pi-WAM. Just lookup the literature
about pi-calulus. I published it a few days ago,
its part of 2.2.4 released already:

Parallel -C-WAM: An Interleaved Synchronous Emulator
https://medium.com/2989/0196089e143a

Whats your point, Rossy Boy? Except you post pretend
nonsense not knowing what you are doing?

Bye

Ross Finlayson schrieb:

No, troll, these are serial algorithms their optimized forms. >>>>>>> >
Normal sorts of forms, ....


Yeah, everybody already figured out "interpreters" and
"programs" and "spawning".

Go spawn yourself.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in >>>>>>>> > one "run", i.e. a stall-less, branch-less, call-less list of >>>>>>>> less than
a few or less than a few dozens or less than a few hundreds >>>>>>>> > instructions, the results "findings" in data and corresponding >>>>>>>> > "matchings" of expressions, that runs in less than one

microsecond.

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, >>>>>>>> that
the indicators of the above as "positive presence" then is to >>>>>>>> make
for that the adjustments to the offsets and extents and the >>>>>>>> shifts
is according to those, otherwise no-ops. Then the idea is that a >>>>>>>>

As independent logical thread state, that automatically MIMD
follows?

Whats the problem to solve then?

Bye

Mild Shock schrieb:

Hi,

Rossy Boy is neither Einstein nor Zweistein.
He is not Einstein since Einstein is already dead:

Albert Einstein (1879 - 1955)
https://de.wikipedia.org/wiki/Albert_Einstein

He is also not Zweistein, since he doesn't
understand concepts such as:

- NVIDIA Volta ff. architecture

Also his hands are small, and his breath stinks,
and he lives in the basement of his mother.

Bye

Mild Shock schrieb:

Hi,

Slowly I start understanding numbnuts like
Rossy Boy who don't understand tech, although
they are from UK and not from a 3rd world

country, and also I start understanding morons
like Micro Penis, who are behind a curtain,
and cannot access a lot of tech.

The same holds for SWI Prologs newest campaign
that probably adresses some poor indians that
have neither 5G nor Macs:

1:38:01 The Kyiv keynote disaster
https://www.youtube.com/watch?v=U8goS6B3BbI

Woa! Real time download of Scala, Closure,
etc.. Whats the magic behind that? Some SWI
point of sale, downloading it via its

keyboard and some telephathy module ?

Bye

Mild Shock schrieb:

Hi,

Ride the snake
He's old and his skin is cold
The west is the best
The west is the best
Get here and we'll do the rest
The blue bus is calling us
The blue bus is calling us
Driver, where you taking us?

Apocalypse Now intro: The Doors, The End {1979}
https://www.youtube.com/watch?v=CIrvSJwwJUE

Bye

Hi,

Again I posted everything here:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

The repo says, same time when I posted
the link first time:

This repository was archived by the
owner on Jul 9, 2026. It is now read-only.

Now a USENET user, who had already entitled
himself for a couple of irrational accusations

towards my side, is asking this question:

Chris M. Thomasson schrieb, Jul 24, 2026

Show an outline of what you
need you compute shader to do?

Bravo, thats a delay of a wooping 15 days.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture >>>>>>>>>>>> with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

Huffman was a bright lad who it's said came up with Huffman
coding to get past taking a final arduous exam.

One of my grandfather's name is Huffman, he graduated high-school,
somewhere's a picture of him with his high-school graduating
class, in Cleveland, Ohio, more than a hundred years ago.

Three people graduated high-school that year.

Huffman codes are a fantastic thing and ubiquitous, though,
like any other algorithm & data structure, they depend on
the distribution their best/average/worst cases, since everybody
here knows the general outline of asymptotics and Big O, Little o,
and Theta, about time and space and complexity.


Then, "adaptive algorithms" and "adapative data structures",
start with something like "compression" in usual accounts,
which usually is adaptive in the sense of compressing what's
compressible in a window, and making a histogram of the words
in the alphabet interpreted as a sequence of letters, and making
an optimal sort of Huffman-coding for entropy-coding for that,
since Huffman-coding is naturally optimal for a particular distribution,
that like power laws can be found in anything,
that one starts with making histograms after combinatorial enumeration,
and thus resulting an encoding into a language of bit-sequences
with the prefix-property that un-ambiguously makes for compressing
the compressible data.

Then the "adaptive" part of that is
periodically throwing that away and starting another.
One might aver that a more contextually-advised account
studies the entire corpus, for things like CCITT G4, JPEG,
JBIG, Deflate, and any account of entropy-coding or compression,
or Morse code, with of course both Huffman-coding and arithmetic-coding,
since while Huffman-coding is obvious to everybody,
for a while some people thought arithmetic coding had patents,
which now are gone away, leaving all the above mentioned
and MPEG-4 also the "un-encumbered".

The notion of "summary statistics" and "order statistics" though,
in concrete mathematics of course is simple and clear.


Here's an article I read the other day about Huffman-coding,
I found it very insightful and quite enjoyable.

https://fgiesen.wordpress.com/2026/06/21/pivco-huffman-merge-operations/




Shut Up

"It easily scales both up and down
with the capabilities of the target machine."

-
https://fgiesen.wordpress.com/2026/06/21/pivco-huffman-merge-operations/

"A vulnerability in the Xpdf implementation of JBIG2, re-used in Apple's
iOS phone operating software, was used by the Pegasus spyware to
implement a zero-click attack on iPhones by constructing an emulated
computer architecture inside a JBIG2 stream. Apple fixed this
"FORCEDENTRY" vulnerability in iOS 14.8 in September 2021."



--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Since we have a good flow, and since NPUs
share the same system memory, and possibly a
lot of other traits as well with the GPU in
libary(edge/furryhaze), we just developed.

The idea here is to do first some off Dogelog
experiments and then create a library that
provides npu_exec/2 for pi-WAM code, the analogue
to gpu_exec/2. A name suggestion would be:

- edge/ironpaw.p
The new Prolog library

The NPU will be clearly underutilized when only
doing scalar, not sure whether this is even
permitted. But in the long run it is planned
that pi-WAM will have vector and matrix traits
anyways. Here is an example goal can be run

with matrix and quantization traits:

?- [X,Y] ins 0..3, Z is X*2+Y*3+4, T is X*3-Y*2-1

These traits will demand some CPU, GPU
and NPU translation. If we keep these traits
simple, we might indeed arrive at concrete
realization from the same abstract machine

LoL, ironpaw the little brother of ironfist.

Bye

Mild Shock schrieb:

Hi,

Just downloading some other person's code

I didn't do that, I wrote Hack VM for pi-WAM
from scratch, over the last 4 weeks. I came
back from holidays on end of June 2026, and now

we have end of July 2026. But its only possible
because the instruction set is very smal, like
ca. 8 functions and ca. 8 modes and ca. 8 conditions,

so its ca. 8 x 8 x 8 = 512 opcodes, each has an
A parameter and a D parameter simultaneously.
It has currently the following CPU backends:

-a- Now supports interleaved synchronous emulation.
-a- Now supports warp parallelism via Java platform threads.
-a- Now supports warp parallelism via Python system threads.
-a- Now supports warp parallelism via JavaScript worker threads.
-a- Note: For Python free threads are not yet fully tested.
-a- Note: For JavaScript web workers are not yet fully tested.

https://www.dogelog.ch/typtab/doclet/book/14_install/05_notes22/110_224.html

But frankly I came to encounter Hack not from
the usual university curriculum web resources,
but indirectly through a post about a Prolog

emulation of Hack, using constrained horn clauses (CHC):

Verifying Nand2Tetris Assembly
https://www.philipzucker.com/nand2tetris-chc/

The binary encoding is currently that the functions,
modes and conditions eat up a nibble (4-bit), in
total 12-bit, which I use then 10-bit for A parameter

and 10-bit for D parameter. I used AI freemium, Codex
by ChatGPT from within IntelliJ to do some fragment
code translations automatically from Java to JavaScript

or from JavaScript to Python.

Have Fun!

Bye

Mild Shock schrieb:

Hi,

Who exactly is the thief? Does this person
have stats in the Rogue class in dungeons
and dragons?

The conspiracy theory of a stealing of Torso VDBE,
by Rossy Boy, is probably a result of complete
ignorance of the Hack ecosystem.

Hack is a very popular computer science project,
with a couple of subprojects in hardware and
software. It goes also by the name Nand to Tetris,

and is programming language agnositic. You can do
Hack experiments in any programming language, be
it BASIC, ADA or Rust. Nobody cares.

The gist are projects like here, first to
educate yourself about Hack:

https://www.nand2tetris.org/course

And then to use Hack in different contexts:

https://www.nand2tetris.org/copy-of-talks

For didactic purposes, I used Hack for my WebGPU
experiment. I didn't even take a look at Torso
VDBE, why should I? Hack is nicely documented,

has even a book, and fusing the two 16-bit
instruction types A and D, into a single 32-bit
instruction stream, is nowhere patented.

Bye

Mild Shock schrieb:

Hi,

I don't use Rust, you are crazy. First of
all the parallel simulator is 100% written
in Prolog, should also run in ISO Prolog,

enhanced by a library(lists). Second I only
mentioned that WebGPU / WGSL, the language
there has a Rust inspired language.

Its not Rust. Whats wrong with you? Why do
you adress your weariness of life to me.
I am neither thief, nor can I help you

with your frustration, and histeric outbursts.
Maybe just be a man and jump off a bridge, idiot.
Or tame your frustration, usenet is not for

you alone, your stupid asshole.

Bye

Ross Finlayson schrieb:

https://www.theregister.com/databases/2026/07/29/after-rewriting-sqlite-in-rust-turso-turns-its-sights-on-postgres/5279835

I don't much care about Rust.

.. gibberish ..

Thief.

Mild Shock schrieb:

Hi,

Rossy Boys tears could cool a data center,
he thinks there exists no literature about
serial algorithms of parallel stuff, and

he also thinks normal forms lead to optimizing
something. LoL, what a utter bullshit. I did
alreay a serial implementation of a parallel

simulation of my pi-WAM. Just lookup the literature
about pi-calulus. I published it a few days ago,
its part of 2.2.4 released already:

Parallel -C-WAM: An Interleaved Synchronous Emulator
https://medium.com/2989/0196089e143a

Whats your point, Rossy Boy? Except you post pretend
nonsense not knowing what you are doing?

Bye

Ross Finlayson schrieb:

No, troll, these are serial algorithms their optimized forms.

Normal sorts of forms, ....


Yeah, everybody already figured out "interpreters" and
"programs" and "spawning".

Go spawn yourself.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of

less than

a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one microsecond. >>>>>

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, that >>>>> -a> the indicators of the above as "positive presence" then is to make >>>>> -a> for that the adjustments to the offsets and extents and the shifts >>>>> -a> is according to those, otherwise no-ops. Then the idea is that a >>>>>

As independent logical thread state, that automatically MIMD follows? >>>>>
Whats the problem to solve then?

Bye

Mild Shock schrieb:

Hi,

Rossy Boy is neither Einstein nor Zweistein.
He is not Einstein since Einstein is already dead:

Albert Einstein (1879 - 1955)
https://de.wikipedia.org/wiki/Albert_Einstein

He is also not Zweistein, since he doesn't
understand concepts such as:

- NVIDIA Volta ff. architecture

Also his hands are small, and his breath stinks,
and he lives in the basement of his mother.

Bye

Mild Shock schrieb:

Hi,

Slowly I start understanding numbnuts like
Rossy Boy who don't understand tech, although
they are from UK and not from a 3rd world

country, and also I start understanding morons
like Micro Penis, who are behind a curtain,
and cannot access a lot of tech.

The same holds for SWI Prologs newest campaign
that probably adresses some poor indians that
have neither 5G nor Macs:

1:38:01 The Kyiv keynote disaster
https://www.youtube.com/watch?v=U8goS6B3BbI

Woa! Real time download of Scala, Closure,
etc.. Whats the magic behind that? Some SWI
point of sale, downloading it via its

keyboard and some telephathy module ?

Bye

Mild Shock schrieb:

Hi,

Ride the snake
He's old and his skin is cold
The west is the best
The west is the best
Get here and we'll do the rest
The blue bus is calling us
The blue bus is calling us
Driver, where you taking us?

Apocalypse Now intro: The Doors, The End {1979}
https://www.youtube.com/watch?v=CIrvSJwwJUE

Bye

Hi,

Again I posted everything here:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

The repo says, same time when I posted
the link first time:

This repository was archived by the
owner on Jul 9, 2026. It is now read-only.

Now a USENET user, who had already entitled
himself for a couple of irrational accusations

towards my side, is asking this question:

Chris M. Thomasson schrieb, Jul 24, 2026

Show an outline of what you
need you compute shader to do?

Bravo, thats a delay of a wooping 15 days.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

On could believe the AI boom is a kind of
Charles Darvin Galapagos Island Evolution
Trick of repurposing FFT hardware.

But this is of course not true, HPC, high
performance computing, has already defined
level 3 ops years ago.

But look at this rabit hole of Ryzen AI 7 350
NPU design, which is a stripped down Xilinx,
stripped of exotic FFT features:

Getting peak TOPS on a Ryzen AI 7 350 NPU https://destevez.net/2026/05/getting-peak-tops-on-a-ryzen-ai-7-350-npu/

But the core feature, very long instruction
word (VLIW) engines, with hardware accelerated
GEMMs, scattered in grids of ASIC tiles,

connected by DMA and NoC, is even not very
specific to AMD, you find it also in Snapdragon /
Qualcomm SoCs for AI Laptops.

Bye

P.S.: My brain playing tricks, why should I
name a library(ironpaw) ? From the same
article above. Maybe WebNN is easier to use?

"mlir-aie contains a Python framework called
IRON that generates LLVM MLIR code representing
a workload that runs on the NPU, including the
code that runs on each compute tile processor

and the configuration of DMAs and other hardware.
Kernels for the compute tile processor can be
written in C++ and compiled either with the
open-source llvm-aie Peano compiler, which is

a fork of LLVM that adds support for the Xilinx
AI engine processors, or with the closed-source
Xilinx CHESS compiler, which is included in Vitis.
In simple cases the kernels can also be directly

written in Python with IRON."

Getting peak TOPS on a Ryzen AI 7 350 NPU https://destevez.net/2026/05/getting-peak-tops-on-a-ryzen-ai-7-350-npu/

Mild Shock schrieb:

Hi,

Mostlikely for high performance computing |a la,
the Actor/Erlang model is dead, they might rely
on MPMC (Multiple Producer, Multiple Consumer)

queue entities separate from the threads. The
ISO Prolog multi-threading support had also such
threads. But besides that was also Actor/Erlang

leaning in practice, like SWI, where threads
have some default queues. So an actor is basically
a Thread and Mailbox conflation. While a MPMC queue

is a kind of separate Mailbox, where multiple
"actors" can read from and write from. A kind of
localized Linda Tuple store.

Which Programming language did adopted the
non-Actor pi-calculus model? Right golang
with its channels.

Bye

Mild Shock schrieb:

Hi,

Usual question:

Why implement both pre-emptive threading

AND cooperative tasks/engines?

I had implemented the ISO proposal in formerly Jekejeke
Prolog, you find the ISO proposal here:

ISO/IEC DTR 13211rCo5:2007
Prolog multi-threading support
https://logtalk.org/plstd/threads.pdf

But the ISO proposal doesn't match modern WebGPU APIs,
where your logical threads can live remotely in a dedicated GPU
in the VRAM there, and where you would have launch

parameters that say: Hey please run 4096 compute
shaders for me, that have independet thread state. Using
cooperative multi-tasking as the orchestrator works well.

Bye

Mild Shock schrieb:

Hi,

Why does this Lama have a red pyjama.
Oh, its a baby Lama. Its still in the cradle
and needs some training:

RedPajama-Data-v2
https://github.com/togethercomputer/RedPajama-Data

But then Andrej Karpathy recently showed
GPT-2 training on rented GPUs for less
than 100 USD in less then 2 hours.

So where do these grown up Lamas go.
Well Georgi Gerganov prefered C++/C
when he shouted Llama Llama Red Pyjama.

But you also find WebLLM, wrapping the
underlying C++/C GPU interface via the
W3C standard WebGPU / WGSL, with JavaScript:

In-Browser LLM Inference Engine
https://webllm.mlc.ai/

My experience with WebLLM 6 months
ago on an iPad Pro 2024, still a little early
stage performance and robustness.

But hey hardware of AI mobile iGPUs is
still evolving, and AI laptop, AI smartphones
and AI tablets, will soon feature Chinese

hardware such some new Kirin AI in 2027.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Tablets and phone are more annoying to
use with WebGPU. The usual browsers don't
have a Chrome DevTools panel integrated,

so that one could do JavaScript Debugging
directly on the device. Instead one has to
use a desktop machine, and connect the

device via UBS-C , and start a Chrome
Browser there . And then start a Chrome
DevTools panel alone, that is pair with

the device, via UBS-C cable. So this way
I already see where it crashes on the
tablets and phone:

await output.mapAsync(GPUMapMode.READ)
Unhandled Promise Rejection: OperationError

The above is the error that one can re-produce
already here with this test:

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

Not sure what exactly happens. Maybe
a form of timeout or device lost, that the
primitive HTML / JavaScript doesn't handle

gracefully yet. Maybe redimensioning the
test, so that it consumes less time would
help. Who knows? Will see. For production

use of a GPU integration I have to anyway
provide work slicing it seems.

Bye

Mild Shock schrieb:

Hi,

Why does this Lama have a red pyjama.
Oh, its a baby Lama. Its still in the cradle
and needs some training:

RedPajama-Data-v2
https://github.com/togethercomputer/RedPajama-Data

But then Andrej Karpathy recently showed
GPT-2 training on rented GPUs for less
than 100 USD in less then 2 hours.

So where do these grown up Lamas go.
Well Georgi Gerganov prefered C++/C
when he shouted Llama Llama Red Pyjama.

But you also find WebLLM, wrapping the
underlying C++/C GPU interface via the
W3C standard WebGPU / WGSL, with JavaScript:

In-Browser LLM Inference Engine
https://webllm.mlc.ai/

My experience with WebLLM 6 months
ago on an iPad Pro 2024, still a little early
stage performance and robustness.

But hey hardware of AI mobile iGPUs is
still evolving, and AI laptop, AI smartphones
and AI tablets, will soon feature Chinese

hardware such some new Kirin AI in 2027.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Looking at the floor plan of a NPU:

Getting peak TOPS on a Ryzen AI 7 350 NPU https://destevez.net/2026/05/getting-peak-tops-on-a-ryzen-ai-7-350-npu/

It seems to me comms between tiles takes
at least Manhattan Distance or L1 Norm time,
if there is no comms congestion

But how does a packet travel? This way:

+----E
|
|
S

Or this way, from start S to end E:

+-E
+
+
S

And what does the chip do if there is
traffic congestion? Some papers are
here, possibly an old problem giving

that processor "cubes" are nothing new.
But a "cube" would be 3D and not 2D.
This paper is old from 2007 or so:

Routing Algorithms for 2D NoC Architectures http://cva.stanford.edu/classes/ee382c/research/2DRouting.pdf

Bye

Mild Shock schrieb:

Hi,

Tablets and phone are more annoying to
use with WebGPU. The usual browsers don't
have a Chrome DevTools panel integrated,

so that one could do JavaScript Debugging
directly on the device. Instead one has to
use a desktop machine, and connect the

device via UBS-C , and start a Chrome
Browser there . And then start a Chrome
DevTools panel alone, that is pair with

the device, via UBS-C cable. So this way
I already see where it crashes on the
tablets and phone:

await output.mapAsync(GPUMapMode.READ)
Unhandled Promise Rejection: OperationError

The above is the error that one can re-produce
already here with this test:

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

Not sure what exactly happens. Maybe
a form of timeout or device lost, that the
primitive HTML / JavaScript doesn't handle

gracefully yet. Maybe redimensioning the
test, so that it consumes less time would
help. Who knows? Will see. For production

use of a GPU integration I have to anyway
provide work slicing it seems.

Bye

Mild Shock schrieb:

Hi,

Why does this Lama have a red pyjama.
Oh, its a baby Lama. Its still in the cradle
and needs some training:

RedPajama-Data-v2
https://github.com/togethercomputer/RedPajama-Data

But then Andrej Karpathy recently showed
GPT-2 training on rented GPUs for less
than 100 USD in less then 2 hours.

So where do these grown up Lamas go.
Well Georgi Gerganov prefered C++/C
when he shouted Llama Llama Red Pyjama.

But you also find WebLLM, wrapping the
underlying C++/C GPU interface via the
W3C standard WebGPU / WGSL, with JavaScript:

In-Browser LLM Inference Engine
https://webllm.mlc.ai/

My experience with WebLLM 6 months
ago on an iPad Pro 2024, still a little early
stage performance and robustness.

But hey hardware of AI mobile iGPUs is
still evolving, and AI laptop, AI smartphones
and AI tablets, will soon feature Chinese

hardware such some new Kirin AI in 2027.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

As easy as queues and FIFO objects might
sound. They don't like congestion. NACK for
retransmission might double the Manhattan Distance:

You have not only start
S to end E communication:

+----E
|
|
S

You might also have ACK or NACK
from E or midpoints back to S:

S'
+
+
E'

Ok, I made that up, I have no idea what a flit is,
when the author wrote this here:

"Packet flits are held in the FIFO which can
be used to determine back pressure. Dropping flits
in a NoC may not be possible since these
architectures may not provide an end-to-end
protocol for retransmission."

Routing Algorithms for 2D NoC Architectures http://cva.stanford.edu/classes/ee382c/research/2DRouting.pdf

Bye

Mild Shock schrieb:

Hi,

Looking at the floor plan of a NPU:

Getting peak TOPS on a Ryzen AI 7 350 NPU https://destevez.net/2026/05/getting-peak-tops-on-a-ryzen-ai-7-350-npu/

It seems to me comms between tiles takes
at least Manhattan Distance or L1 Norm time,
if there is no comms congestion

But how does a packet travel? This way:

+----E
|
|
S

Or this way, from start S to end E:

-a-a +-E
-a +
-a+
S

And what does the chip do if there is
traffic congestion? Some papers are
here, possibly an old problem giving

that processor "cubes" are nothing new.
But a "cube" would be 3D and not 2D.
This paper is old from 2007 or so:

Routing Algorithms for 2D NoC Architectures http://cva.stanford.edu/classes/ee382c/research/2DRouting.pdf

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

On 8/1/26 05:19, Mild Shock wrote:

Hi,

Tablets and phone are more annoying to
use with WebGPU. The usual browsers don't
have a Chrome DevTools panel integrated,

so that one could do JavaScript Debugging
directly on the device. Instead one has to
use a desktop machine, and connect the

device via UBS-C , and start a Chrome
Browser there . And then start a Chrome
DevTools panel alone, that is pair with

the device, via UBS-C cable. So this way
I already see where it crashes on the
tablets and phone:

await output.mapAsync(GPUMapMode.READ)
Unhandled Promise Rejection: OperationError

The above is the error that one can re-produce
already here with this test:

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

Not sure what exactly happens. Maybe
a form of timeout or device lost, that the
primitive HTML / JavaScript doesn't handle

gracefully yet. Maybe redimensioning the
test, so that it consumes less time would
help. Who knows? Will see. For production

use of a GPU integration I have to anyway
provide work slicing it seems.

Bye

I know that you are Hanson and I know that you are a cocksucker.

My question is, though, do you know, do, vomit, defecate, smell,
evaporate, sweat, ooze in, ooze out, fart, see, hear, sense, taste, and GESTATE anything other than programming?

Programming is just a tool, you know. It is nothing by itself worth even mentioning.


--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

WebGPU and WebGL are two different things. I explained
that towards you already like 3-5 times.

Of course we can make a special texture to handle it.

You still don't understand that I am using WebGPU,
and not WebGL. WebGPU has three improvements,
that from your talking are missing in WebGL?

- It has compute shaders
- It has arrays
- It has structs
- What else?

I didn't use structs in my example, although Gemini
nearly forced me to use structs. But you could
use a struct with fields and some of these arrays

to represent a queue. But here in this example
that is open source, I only used flat arrays. I
nowhere needed to abuse textures to store something:

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

You can study the source code, the arrays have
CUDA inspired binding annotations but are not
CUDA but rather WGSL:

Hack VM as a Compute Shader in WGSL
@group(0) @binding(0) var<storage, read> code: array<i32>;
@group(0) @binding(1) var<storage, read_write> state: array<i32>; https://github.com/Jean-Luc-Picard-2021/gigabudget/blob/main/course/example63/boot.mjs

You can say whether a buffer is read, or read_write.
Buffers can be transfered from CPU to GPU, before
running commands, and transfered back from GPU to

CPU after running commands. The use case that
you find on GitHub uses both. Namely also fetching
results via a buffer, to then show them in

the HTML page. Shouldn't be much a problem to
run the example at home locally, all you need is
a HTTPS server. But the example is not yet queues,

but it already shows the foundation, which is WebGPU
with its language WGSL and not WebGL with its language
GLSL. These are two different things.

I explained that towards you already like 3-5 times.

Bye

Chris M. Thomasson schrieb:

On 8/1/2026 5:22 AM, Mild Shock wrote:

Hi,

As easy as queues and FIFO objects might
sound. They don't like congestion. NACK for
retransmission might double the Manhattan Distance:

[...]

You are going to need a place to allocate nodes in the compute

shader. Of course we can make a special texture to handle it. But, we
need to strive to avoid a wait condition. I don't want a compute shader
to spin. Yes, CAS can be used, but, try to make it be used as a "state machine", where the transitions from states are atomic. Try to avoid it
making a loop, where we loop on failure.


Mild Shock schrieb:

Hi,

As easy as queues and FIFO objects might
sound. They don't like congestion. NACK for
retransmission might double the Manhattan Distance:

You have not only start
S to end E communication:

+----E
|
|
S

You might also have ACK or NACK
from E or midpoints back to S:

-a-a S'
-a +
-a+
E'

Ok, I made that up, I have no idea what a flit is,
when the author wrote this here:

"Packet flits are held in the FIFO which can
be used to determine back pressure. Dropping flits
in a NoC may not be possible since these
architectures may not provide an end-to-end
protocol for retransmission."

Routing Algorithms for 2D NoC Architectures http://cva.stanford.edu/classes/ee382c/research/2DRouting.pdf

Bye

Mild Shock schrieb:

Hi,

Looking at the floor plan of a NPU:

Getting peak TOPS on a Ryzen AI 7 350 NPU
https://destevez.net/2026/05/getting-peak-tops-on-a-ryzen-ai-7-350-npu/

It seems to me comms between tiles takes
at least Manhattan Distance or L1 Norm time,
if there is no comms congestion

But how does a packet travel? This way:

+----E
|
|
S

Or this way, from start S to end E:

-a-a-a +-E
-a-a +
-a-a+
S

And what does the chip do if there is
traffic congestion? Some papers are
here, possibly an old problem giving

that processor "cubes" are nothing new.
But a "cube" would be 3D and not 2D.
This paper is old from 2007 or so:

Routing Algorithms for 2D NoC Architectures
http://cva.stanford.edu/classes/ee382c/research/2DRouting.pdf

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

It has textures to work with in the pipeline.

Hi,

Why would I use text inside my compute shader.
Could you tell me. The Hack VM doesn't do
textures. You are confused. There is nothing

about textures here:

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

You can read the text , it says nowhere
consume or produce textures. Its not a rendering
application. I use the compute shader to run Prolog:

"At the end of 2025 we acquired a couple of
AI Laptops , that were still cheap, since
RAM prices had not yet rocketed. The intend
was to tap into the Copilot+ certified hardware,
and shave off some of the TOPS to do Prolog
inferencing. Amazingly our -C-WAM can
churn 11.4 GIGA LIPS.

GPUs have evolved form lock-step to independent
thread scheduling. This made it possible to
port the Hack VM variant, that forms the basis
for our -C-WAM, to WebGPU computer shaders.
Using NUM_SHADERS = 4096 we could produce
11.4 Giga Lips on a Ryzen AI 7 350 w/ Radeon 860M."

Bye

Mild Shock schrieb:

Hi,

WebGPU and WebGL are two different things. I explained
that towards you already like 3-5 times.

Of course we can make a special texture to handle it.

You still don't understand that I am using WebGPU,
and not WebGL. WebGPU has three improvements,
that from your talking are missing in WebGL?

- It has compute shaders
- It has arrays
- It has structs
- What else?

I didn't use structs in my example, although Gemini
nearly forced me to use structs. But you could
use a struct with fields and some of these arrays

to represent a queue. But here in this example
that is open source, I only used flat arrays. I
nowhere needed to abuse textures to store something:

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

You can study the source code, the arrays have
CUDA inspired binding annotations but are not
CUDA but rather WGSL:

Hack VM as a Compute Shader in WGSL
@group(0) @binding(0) var<storage, read> code: array<i32>;
@group(0) @binding(1) var<storage, read_write> state: array<i32>; https://github.com/Jean-Luc-Picard-2021/gigabudget/blob/main/course/example63/boot.mjs

You can say whether a buffer is read, or read_write.
Buffers can be transfered from CPU to GPU, before
running commands, and transfered back from GPU to

CPU after running commands. The use case that
you find on GitHub uses both. Namely also fetching
results via a buffer, to then show them in

the HTML page. Shouldn't be much a problem to
run the example at home locally, all you need is
a HTTPS server. But the example is not yet queues,

but it already shows the foundation, which is WebGPU
with its language WGSL and not WebGL with its language
GLSL. These are two different things.

I explained that towards you already like 3-5 times.

Bye

Chris M. Thomasson schrieb:

On 8/1/2026 5:22 AM, Mild Shock wrote:

Hi,

As easy as queues and FIFO objects might
sound. They don't like congestion. NACK for
retransmission might double the Manhattan Distance:

[...]

You are going to need a place to allocate nodes in the compute

shader. Of course we can make a special texture to handle it. But, we
need to strive to avoid a wait condition. I don't want a compute shader
to spin. Yes, CAS can be used, but, try to make it be used as a "state machine", where the transitions from states are atomic. Try to avoid it making a loop, where we loop on failure.

Mild Shock schrieb:

Hi,

As easy as queues and FIFO objects might
sound. They don't like congestion. NACK for
retransmission might double the Manhattan Distance:

You have not only start
S to end E communication:

+----E
|
|
S

You might also have ACK or NACK
from E or midpoints back to S:

-a-a-a S'
-a-a +
-a-a+
E'

Ok, I made that up, I have no idea what a flit is,
when the author wrote this here:

"Packet flits are held in the FIFO which can
be used to determine back pressure. Dropping flits
in a NoC may not be possible since these
architectures may not provide an end-to-end
protocol for retransmission."

Routing Algorithms for 2D NoC Architectures
http://cva.stanford.edu/classes/ee382c/research/2DRouting.pdf

Bye

Mild Shock schrieb:

Hi,

Looking at the floor plan of a NPU:

Getting peak TOPS on a Ryzen AI 7 350 NPU
https://destevez.net/2026/05/getting-peak-tops-on-a-ryzen-ai-7-350-npu/

It seems to me comms between tiles takes
at least Manhattan Distance or L1 Norm time,
if there is no comms congestion

But how does a packet travel? This way:

+----E
|
|
S

Or this way, from start S to end E:

-a-a-a +-E
-a-a +
-a-a+
S

And what does the chip do if there is
traffic congestion? Some papers are
here, possibly an old problem giving

that processor "cubes" are nothing new.
But a "cube" would be 3D and not 2D.
This paper is old from 2007 or so:

Routing Algorithms for 2D NoC Architectures
http://cva.stanford.edu/classes/ee382c/research/2DRouting.pdf

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

But, I still don't know what you main goal is?

It explicity says "Prolog inferencing" in
this phrase:

shave off some of the TOPS to do Prolog inferencing

It nowhere says draw some fancy stuff into
a Web canvas.

Bye

Mild Shock schrieb:

Hi,

It has textures to work with in the pipeline.

Hi,

Why would I use text inside my compute shader.
Could you tell me. The Hack VM doesn't do
textures. You are confused. There is nothing

about textures here:

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

You can read the text , it says nowhere
consume or produce textures. Its not a rendering
application. I use the compute shader to run Prolog:

"At the end of 2025 we acquired a couple of
AI Laptops , that were still cheap, since
RAM prices had not yet rocketed. The intend
was to tap into the Copilot+ certified hardware,
and shave off some of the TOPS to do Prolog
inferencing. Amazingly our -C-WAM can
churn 11.4 GIGA LIPS.

GPUs have evolved form lock-step to independent
thread scheduling. This made it possible to
port the Hack VM variant, that forms the basis
for our -C-WAM, to WebGPU computer shaders.
Using NUM_SHADERS = 4096 we could produce
11.4 Giga Lips on a Ryzen AI 7 350 w/ Radeon 860M."

Bye

Mild Shock schrieb:

Hi,

WebGPU and WebGL are two different things. I explained
that towards you already like 3-5 times.

Of course we can make a special texture to handle it.

You still don't understand that I am using WebGPU,
and not WebGL. WebGPU has three improvements,
that from your talking are missing in WebGL?

- It has compute shaders
- It has arrays
- It has structs
- What else?

I didn't use structs in my example, although Gemini
nearly forced me to use structs. But you could
use a struct with fields and some of these arrays

to represent a queue. But here in this example
that is open source, I only used flat arrays. I
nowhere needed to abuse textures to store something:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

You can study the source code, the arrays have
CUDA inspired binding annotations but are not
CUDA but rather WGSL:

Hack VM as a Compute Shader in WGSL
@group(0) @binding(0) var<storage, read> code: array<i32>;
@group(0) @binding(1) var<storage, read_write> state: array<i32>;
https://github.com/Jean-Luc-Picard-2021/gigabudget/blob/main/course/example63/boot.mjs


You can say whether a buffer is read, or read_write.
Buffers can be transfered from CPU to GPU, before
running commands, and transfered back from GPU to

CPU after running commands. The use case that
you find on GitHub uses both. Namely also fetching
results via a buffer, to then show them in

the HTML page. Shouldn't be much a problem to
run the example at home locally, all you need is
a HTTPS server. But the example is not yet queues,

but it already shows the foundation, which is WebGPU
with its language WGSL and not WebGL with its language
GLSL. These are two different things.

I explained that towards you already like 3-5 times.

Bye

Chris M. Thomasson schrieb:

On 8/1/2026 5:22 AM, Mild Shock wrote:

Hi,

As easy as queues and FIFO objects might
sound. They don't like congestion. NACK for
retransmission might double the Manhattan Distance:

[...]

You are going to need a place to allocate nodes in the compute

shader. Of course we can make a special texture to handle it. But, we
need to strive to avoid a wait condition. I don't want a compute
shader to spin. Yes, CAS can be used, but, try to make it be used as a
"state machine", where the transitions from states are atomic. Try to
avoid it making a loop, where we loop on failure.


Mild Shock schrieb:

Hi,

As easy as queues and FIFO objects might
sound. They don't like congestion. NACK for
retransmission might double the Manhattan Distance:

You have not only start
S to end E communication:

+----E
|
|
S

You might also have ACK or NACK
from E or midpoints back to S:

-a-a-a S'
-a-a +
-a-a+
E'

Ok, I made that up, I have no idea what a flit is,
when the author wrote this here:

"Packet flits are held in the FIFO which can
be used to determine back pressure. Dropping flits
in a NoC may not be possible since these
architectures may not provide an end-to-end
protocol for retransmission."

Routing Algorithms for 2D NoC Architectures
http://cva.stanford.edu/classes/ee382c/research/2DRouting.pdf

Bye

Mild Shock schrieb:

Hi,

Looking at the floor plan of a NPU:

Getting peak TOPS on a Ryzen AI 7 350 NPU
https://destevez.net/2026/05/getting-peak-tops-on-a-ryzen-ai-7-350-npu/ >>>>
It seems to me comms between tiles takes
at least Manhattan Distance or L1 Norm time,
if there is no comms congestion

But how does a packet travel? This way:

+----E
|
|
S

Or this way, from start S to end E:

-a-a-a +-E
-a-a +
-a-a+
S

And what does the chip do if there is
traffic congestion? Some papers are
here, possibly an old problem giving

that processor "cubes" are nothing new.
But a "cube" would be 3D and not 2D.
This paper is old from 2007 or so:

Routing Algorithms for 2D NoC Architectures
http://cva.stanford.edu/classes/ee382c/research/2DRouting.pdf

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Chris M. Thomasson can ask 100 more questions.
I will happily answer them. But maybe I should
make a Wiki to explain the ever same things:

But, I still don't know what you main goal is?

The goal is "Prolog inferencing"

It has textures to work with in the pipeline.

I don't need textures for "Prolog inferencing"

98 more questions to go, don't give up!

Bye

Taskfreak schrieb:

On 8/1/26 05:19, Mild Shock wrote:

Hi,

Tablets and phone are more annoying to
use with WebGPU. The usual browsers don't
have a Chrome DevTools panel integrated,

so that one could do JavaScript Debugging
directly on the device. Instead one has to
use a desktop machine, and connect the

device via UBS-C , and start a Chrome
Browser there . And then start a Chrome
DevTools panel alone, that is pair with

the device, via UBS-C cable. So this way
I already see where it crashes on the
tablets and phone:

await output.mapAsync(GPUMapMode.READ)
Unhandled Promise Rejection: OperationError

The above is the error that one can re-produce
already here with this test:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

Not sure what exactly happens. Maybe
a form of timeout or device lost, that the
primitive HTML / JavaScript doesn't handle

gracefully yet. Maybe redimensioning the
test, so that it consumes less time would
help. Who knows? Will see. For production

use of a GPU integration I have to anyway
provide work slicing it seems.

Bye

I know that you are Hanson and I know that you are a cocksucker.

My question is, though, do you know, do, vomit, defecate, smell,
evaporate, sweat, ooze in, ooze out, fart, see, hear, sense, taste, and GESTATE anything other than programming?

Programming is just a tool, you know. It is nothing by itself worth even mentioning.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Ok, following the instructions here:

npm install webgpu
https://github.com/dawn-gpu/node-webgpu

I can now run webgpu also from CLI:

node.exe dogelog.mjs

Dogelog Spieler 2.2.5, Node, JavaScript 26.4.0
(c) 1985-2026, XLOG Technologies AG, Schweiz

?- ensure_loaded(library(edge/furryhaze)).
true.

?- between(1,3,_), time(expedite((between(1,100,_),
between(1,100,_), between(1,100,_)), [size(4096)])), fail.
% Zeit 1037.994 ms, GC 0.000 ms, Lips 111 k
% Zeit 1091.131 ms, GC 0.000 ms, Lips 106 k
% Zeit 1045.274 ms, GC 0.000 ms, Lips 110 k
fail.

Same benchmark result as in the browser.
Now I can rent a bigger GPU by the hour
and do some easy CLI testing.

LoL

Bye

Mild Shock schrieb:

Hi,

Chris M. Thomasson can ask 100 more questions.
I will happily answer them. But maybe I should
make a Wiki to explain the ever same things:

But, I still don't know what you main goal is?

The goal is "Prolog inferencing"

It has textures to work with in the pipeline.

I don't need textures for "Prolog inferencing"

98 more questions to go, don't give up!

Bye

Taskfreak schrieb:

On 8/1/26 05:19, Mild Shock wrote:

Hi,

Tablets and phone are more annoying to
use with WebGPU. The usual browsers don't
have a Chrome DevTools panel integrated,

so that one could do JavaScript Debugging
directly on the device. Instead one has to
use a desktop machine, and connect the

device via UBS-C , and start a Chrome
Browser there . And then start a Chrome
DevTools panel alone, that is pair with

the device, via UBS-C cable. So this way
I already see where it crashes on the
tablets and phone:

await output.mapAsync(GPUMapMode.READ)
Unhandled Promise Rejection: OperationError

The above is the error that one can re-produce
already here with this test:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

Not sure what exactly happens. Maybe
a form of timeout or device lost, that the
primitive HTML / JavaScript doesn't handle

gracefully yet. Maybe redimensioning the
test, so that it consumes less time would
help. Who knows? Will see. For production

use of a GPU integration I have to anyway
provide work slicing it seems.

Bye

I know that you are Hanson and I know that you are a cocksucker.

My question is, though, do you know, do, vomit, defecate, smell,
evaporate, sweat, ooze in, ooze out, fart, see, hear, sense, taste,
and GESTATE anything other than programming?

Programming is just a tool, you know. It is nothing by itself worth
even mentioning.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

If only the fucking moron Chris M. Thomasson would
stop spamming his nonsense, he doesn't listen at
all. Problem, he cannot read, he knows nothing.

Its very common that compute shaders can block,
when they are used for General Purpose computation
on GPUs (GPGPU). If only he would pull out his

finger from his asshole, and stop thinking in his
WebGL legacy code stash nonsense. Even the
Cerebras Waver has blocking:

"Cerebras Software Language (CSL), send_color
and recv_color are parameters passed to tile
programs to manage data routing and virtual
channels (called colors) across processing
elements (PEs) on the wafer

Yes, both send and receive operations can block
on a Cerebras Processing Element (PE), primarily
due to the system's hardware-enforced backpressure
mechanism. Because the Cerebras Wafer-Scale Engine
(WSE) relies on a fine-grained,

dataflow-driven architecture, blocking prevents
data loss when hardware resources are
fully saturated."

Blocking and Unblocking https://sdk.cerebras.ai/computing-with-cerebras#blocking-and-unblocking

Chris M. Thomasson is an annoyance and an idiot.
He is a total waste of time. And represents those
people who cannot use their brain.

Bye

Chris M. Thomasson schrieb:

On 8/1/2026 5:47 PM, Mild Shock wrote:

Hi,

Chris M. Thomasson can ask 100 more questions.
I will happily answer them. But maybe I should
make a Wiki to explain the ever same things:

But, I still don't know what you main goal is?

The goal is "Prolog inferencing"

It has textures to work with in the pipeline.

I don't need textures for "Prolog inferencing"

98 more questions to go, don't give up!

[...]

Fwiw, I have several compute shaders that do what I want. Mainly

building vector fields, etc.... And yes I use textures for some input
and output, uniforms mainly for the settings, etc. Just, make sure to
code things up to a point where your compute shader never needs to wait
for something... Think of striving for wait-free algorithms.

For instance, this is 100% wait free.

void add_hit(ct_plane2d plane, vec2 p, vec3 weight)
{
vec2 uv = ct_plane2d_unproject(plane, p);
ivec2 px = ivec2(uv * u_resolution);

if (px.x >= 0 && px.x < int(u_resolution.x) &&
px.y >= 0 && px.y < int(u_resolution.y))
{
imageAtomicAdd(accum_r, px, weight.r);
imageAtomicAdd(accum_g, px, weight.g);
imageAtomicAdd(accum_b, px, weight.b);
imageAtomicAdd(accum_hits, px, 1.0f);
}
}

Notice how I separated my accumulation buffer into different textures?

layout(binding = 0, r32f) uniform coherent image2D accum_r;
layout(binding = 1, r32f) uniform coherent image2D accum_g;
layout(binding = 2, r32f) uniform coherent image2D accum_b;
layout(binding = 3, r32f) uniform coherent image2D accum_hits; //

alpha / hit counter

Works great and runs really fast.

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly. https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Chris M. Thomasson schrieb:

Strive to never make a compute shader wait
on something, like an empty condition of a queue, stack.

You are such a moron. GPU elasticity was
already invented in 2008 with CUDA. I posted
this quote already:

"CUDArao TEChNOLOGY UNLOCkS ThE
POWER OF TESLA MANY-CORE PROCESSORS

The CUDA C compiler simplifies many-core
programming by enabling code development
in a high-level language and optimizing code
to run on systems without knowledge of how

many cores are in the hardware. CUDA
applications automatically take advantage of
more cores or fewer cores in a system, so
they can scale from entry-level notebook

GPUs to high end GPUs in technical workstations
and further into racks of GPUs in data centers.
This allows developers to rCLcode oncerCY and
deploy on a range of systems, as well as scale

forward in time as future GPUs deliver more
performance per watt and more cores per
processor. The benefit for software users is
the opportunity to boost computing performance

simply by adding GPUs or using their existing GPUs in new ways" https://www.nvidia.com/docs/IO/43395/NV_DS_Tesla_S1070_US_Jun08_NV_LR_Final.pdf

Today elasticity is on logical thread aka task level,
not only on "core" level or something. Don't know
exactly what CUDA did back them, maybe only

a submit elasticity, like a time sharing system. Today
you have quite some run elasticity on modern machines,
for your logical threads. Even in budget laptops

like a Ryzen AI 7 350 /w Radeon 850M.

Bye

Mild Shock schrieb:

Hi,

Ok, following the instructions here:

npm install webgpu
https://github.com/dawn-gpu/node-webgpu

I can now run webgpu also from CLI:

node.exe dogelog.mjs

Dogelog Spieler 2.2.5, Node, JavaScript 26.4.0
(c) 1985-2026, XLOG Technologies AG, Schweiz

?- ensure_loaded(library(edge/furryhaze)).
true.

?- between(1,3,_), time(expedite((between(1,100,_),
between(1,100,_), between(1,100,_)), [size(4096)])), fail.
% Zeit 1037.994 ms, GC 0.000 ms, Lips 111 k
% Zeit 1091.131 ms, GC 0.000 ms, Lips 106 k
% Zeit 1045.274 ms, GC 0.000 ms, Lips 110 k
fail.

Same benchmark result as in the browser.
Now I can rent a bigger GPU by the hour
and do some easy CLI testing.

LoL

Bye

Mild Shock schrieb:

Hi,

Chris M. Thomasson can ask 100 more questions.
I will happily answer them. But maybe I should
make a Wiki to explain the ever same things:

But, I still don't know what you main goal is?

The goal is "Prolog inferencing"

It has textures to work with in the pipeline.

I don't need textures for "Prolog inferencing"

98 more questions to go, don't give up!

Bye

Taskfreak schrieb:

On 8/1/26 05:19, Mild Shock wrote:

Hi,

Tablets and phone are more annoying to
use with WebGPU. The usual browsers don't
have a Chrome DevTools panel integrated,

so that one could do JavaScript Debugging
directly on the device. Instead one has to
use a desktop machine, and connect the

device via UBS-C , and start a Chrome
Browser there . And then start a Chrome
DevTools panel alone, that is pair with

the device, via UBS-C cable. So this way
I already see where it crashes on the
tablets and phone:

await output.mapAsync(GPUMapMode.READ)
Unhandled Promise Rejection: OperationError

The above is the error that one can re-produce
already here with this test:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

Not sure what exactly happens. Maybe
a form of timeout or device lost, that the
primitive HTML / JavaScript doesn't handle

gracefully yet. Maybe redimensioning the
test, so that it consumes less time would
help. Who knows? Will see. For production

use of a GPU integration I have to anyway
provide work slicing it seems.

Bye

I know that you are Hanson and I know that you are a cocksucker.

My question is, though, do you know, do, vomit, defecate, smell,
evaporate, sweat, ooze in, ooze out, fart, see, hear, sense, taste,
and GESTATE anything other than programming?

Programming is just a tool, you know. It is nothing by itself worth
even mentioning.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

This was archived on Jul 9, 2026:

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

Still today on Aug 03, 2026, the usenet
community still struggles with the experiment,
doesn't know the meaning and implications,

especially clueless about 4096 shaders and
modern GPU elasticity. Woa! Thats impressive.
Especially Chris M. Thomasson has a still ongoing

hard time with this little WebGPU experiment.

Bye

Mild Shock schrieb:

Hi,

This was archived on Jul 9, 2026:

11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget

Still, Jul 29, Rossy Boy halucinates accusations:

Ross Finlayson schrieb:

.. bla bla goto bla bla ..

Stupid gangster:-a teamsters are a union.

In the trades, not the steals, ....

Woa! Thats now 20 days of brain desease,
and not understanding the meaning and implications.
Even not understand pi-WAM has Hack VM backend.

But its all opensource. Bravo Rossy Boy, you are
champion in brainlessness and lazyness of
a idiot usenet troll.

Bye

Mild Shock schrieb:

Hi,

If any of you guys do not understand what
is meant by or what the implications are:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

Well I wouldn't care less. There are two
outcomes for numb nuts:

- Ignoramus: They don't understand it, but
-a-a they will understand it before they die.

- Ignorabimus: They don't understand it, and
-a-a will never understand it, and they die.

So who cares, its not my problem, you people
are stupid as fuck, and slow as fuck...

Bye

Mild Shock schrieb:

Hi,

Micro penis brain is in constant hiatus.
He can even not detect a trope.

LoL

Bye

Lane W schrieb:

Mild Shock wrote:

Hi,

My mother is worried that I fucked Lane W.
aka Micro Penis mother 24 hours straight.
She was screaming, basically singing all

the arias from operas that Luciano Pavarotti
usually sings. You Lane W. aka Micro Penis
should have heard it, since you

live in the basement of your mothers house.

No, actually remarkably, I don't. According to google I live 433

miles away from her.

Strike!

See, what i said about you was spot on.

What you said about me was generic and incorrect.

You really suck, man.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: sci.physics

Hi,

Or do a YouTube video about:

Standing on the shoulders of giants https://en.wikipedia.org/wiki/Standing_on_the_shoulders_of_giants

Calling people who build software "thieves",
is probably the most philosopher syphilis brain

thing I ever heard in 2026. You should really
jump from a bridge Rossy Boy. I think its over

for you, the lamps have already gone out...

Bye

Mild Shock schrieb:

Hi,

Do a YouTube video about it:

Topic: Tit for Tat, or how I messed up
with an innocent poster, and learnt about FAFO:

#fuckaroundandfindout
https://www.youtube.com/shorts/6ALRRksc72M

You were provable the first idiot, posting
stupid comments into my posts, besides of

course Micro Penis, who is a paid troll.

Have Fun!

Bye

Ross Finlayson schrieb:

For dummies, ....

Mild Shock schrieb:

Hi,

I don't use Rust, you are crazy. First of
all the parallel simulator is 100% written
in Prolog, should also run in ISO Prolog,

enhanced by a library(lists). Second I only
mentioned that WebGPU / WGSL, the language
there has a Rust inspired language.

Its not Rust. Whats wrong with you? Why do
you adress your weariness of life to me.
I am neither thief, nor can I help you

with your frustration, and histeric outbursts.
Maybe just be a man and jump off a bridge, idiot.
Or tame your frustration, usenet is not for

you alone, your stupid asshole.

Bye

Ross Finlayson schrieb:

https://www.theregister.com/databases/2026/07/29/after-rewriting-sqlite-in-rust-turso-turns-its-sights-on-postgres/5279835

I don't much care about Rust.

.. gibberish ..

Thief.

Mild Shock schrieb:

Hi,

Rossy Boys tears could cool a data center,
he thinks there exists no literature about
serial algorithms of parallel stuff, and

he also thinks normal forms lead to optimizing
something. LoL, what a utter bullshit. I did
alreay a serial implementation of a parallel

simulation of my pi-WAM. Just lookup the literature
about pi-calulus. I published it a few days ago,
its part of 2.2.4 released already:

Parallel -C-WAM: An Interleaved Synchronous Emulator
https://medium.com/2989/0196089e143a

Whats your point, Rossy Boy? Except you post pretend
nonsense not knowing what you are doing?

Bye

Ross Finlayson schrieb:

No, troll, these are serial algorithms their optimized forms.

Normal sorts of forms, ....


Yeah, everybody already figured out "interpreters" and
"programs" and "spawning".

Go spawn yourself.

Mild Shock schrieb:

Hi,

You are still chewing on SIMD. LoL

Ross Finlayson schrieb:

Then the idea is that any of those can be found and matched in
one "run", i.e. a stall-less, branch-less, call-less list of less

than

a few or less than a few dozens or less than a few hundreds
instructions, the results "findings" in data and corresponding
"matchings" of expressions, that runs in less than one microsecond.

You cannot make the mental translation that if you have:

Ross Finlayson schrieb:

So, the context then is for register state and stack contents, that
the indicators of the above as "positive presence" then is to make
for that the adjustments to the offsets and extents and the shifts
is according to those, otherwise no-ops. Then the idea is that a

As independent logical thread state, that automatically MIMD follows?

Whats the problem to solve then?

Bye

Mild Shock schrieb:

Hi,

Rossy Boy is neither Einstein nor Zweistein.
He is not Einstein since Einstein is already dead:

Albert Einstein (1879 - 1955)
https://de.wikipedia.org/wiki/Albert_Einstein

He is also not Zweistein, since he doesn't
understand concepts such as:

- NVIDIA Volta ff. architecture

Also his hands are small, and his breath stinks,
and he lives in the basement of his mother.

Bye

Mild Shock schrieb:

Hi,

Slowly I start understanding numbnuts like
Rossy Boy who don't understand tech, although
they are from UK and not from a 3rd world

country, and also I start understanding morons
like Micro Penis, who are behind a curtain,
and cannot access a lot of tech.

The same holds for SWI Prologs newest campaign
that probably adresses some poor indians that
have neither 5G nor Macs:

1:38:01 The Kyiv keynote disaster
https://www.youtube.com/watch?v=U8goS6B3BbI

Woa! Real time download of Scala, Closure,
etc.. Whats the magic behind that? Some SWI
point of sale, downloading it via its

keyboard and some telephathy module ?

Bye

Mild Shock schrieb:

Hi,

Ride the snake
He's old and his skin is cold
The west is the best
The west is the best
Get here and we'll do the rest
The blue bus is calling us
The blue bus is calling us
Driver, where you taking us?

Apocalypse Now intro: The Doors, The End {1979}
https://www.youtube.com/watch?v=CIrvSJwwJUE

Bye

Hi,

Again I posted everything here:

11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget

The repo says, same time when I posted
the link first time:

This repository was archived by the
owner on Jul 9, 2026. It is now read-only.

Now a USENET user, who had already entitled
himself for a couple of irrational accusations

towards my side, is asking this question:

Chris M. Thomasson schrieb, Jul 24, 2026

Show an outline of what you
need you compute shader to do?

Bravo, thats a delay of a wooping 15 days.

Bye

Mild Shock schrieb:

Hi,

Remember when first all local AI was Python
and PyTorch APIs. And then suddently people started
using bare metal C/C++ Code. Here is the story:

How it started:

GPT-J or GPT-J-6B is an open-source large
language model (LLM) developed by EleutherAI
in 2021. As the name suggests, it is a
generative pre-trained transformer model
designed to produce human-like text that
continues from a prompt.
https://www.eleuther.ai/

How it was going [Georgi Gerganov]:

So a few days later comes out the LLaMA, I do
some calculations and I figure out rCLOkay, 65
billion parameters. You probably need about
40 gigs of RAM, with 4-bit quantization. So
this can run on a MacBook. Why not do it?rCY

Why I was able to do it so quickly - basically,
for all that I saw itrCOs pretty much GPT-J architecture
with some modifications, like some extra memorization
layers. ItrCOs minor changes. Basically, again, the
existing code for the GPT-J, I just simply
modified it there, it happened pretty quickly.
https://changelog.com/podcast/532

Georgi Gerganov, Bulgarian, now with Hugging
Face, ggml-cann also running on Chinese AI chips.
ggml Manifesto https://github.com/ggml-org/ggml

Bye

--- Synchronet 3.22a-Linux NewsLink 1.2