From Newsgroup: comp.lang.prolog

Hi,

Is there a silver lining of AI democratization?
With MedGamma I can analyse my own broken ribs.
If only I had a body scanner. I am currently exploring

options of LLM models that I could run on
my new AI soaked AMD Ryzen AI 7 350 laptop.
While Qualcomm spear headed LLM players with

their LM Studio. There is FastFlowLM
that can do Ryzen, and would utilize the NPU. For
example to run a distilled DeepSeek would amount to:

flm run deepseek-r1:8b

And yes there is MedGamma:

MedGemma:4B (Multimodal) Running Exclusively on AMD Ryzenrao AI NPU https://www.youtube.com/watch?v=KWzXZEOcgK4

Bye

Mild Shock schrieb:

Hi,

How it started (released January 20, 2025):

Despite trailing the U.S., China continues to make
significant strides in AI development. A notable example
is the Chinese startup DeepSeek, which released an AI
model named R1. This model demonstrates reasoning
capabilities comparable to leading U.S. models and is
available as open-source software, allowing free use and
commercialization. However, R1 faces issues with censorship,
likely influenced by Chinese regulations. https://en.wikipedia.org/wiki/DeepSeek#Release_history

How its going (released January 21, 2025):

In a strategic move to further solidify its leadership,
President Donald Trump announced the Stargate initiative.
This $500 billion investment, a collaboration between OpenAI,
Oracle, and SoftBank, aims to develop extensive AI
infrastructure across the U.S., including data centers
and energy facilities. The project is expected to create
over 100,000 jobs and enhance the nation's AI capabilities. https://en.wikipedia.org/wiki/The_Stargate_Project

With SoftBank they basically call Japan for help!

LoL

Bye

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.prolog

Hi,

I hope it doesn't turn my laptop into a
frying pan. The thingy had a few hickups
recently, like using 100% CPU and doing

nothing. Maybe IntelliJs fork join framework
was overdoing it. But NPUs are physically
designed for efficient AI math - more

computations per watt, less heat generated.
Lets see. But whats the technology behind
FastFlowLM? It might be a result of:

GAIA: An Open-Source Project from AMD for
Running Local LLMs on Ryzenrao AI. GAIA seems
to be an important piece of the Ryzen AI story.

Initially GAIA wanted to Provide a unified
software stack for Ryzen AI NPUs. But
AMD shifted focus to DirectML integration

with Windows. GAIA Absorbed into AMD's ROCm
Ecosystem, on the other hand XDNA (2024)
AMD's commercial NPU architecture goes

full circle back to Niklaus Wirth:

Hades: fast hardware synthesis tools and a reconfigurable coprocessor https://www.research-collection.ethz.ch/entities/publication/23b3a0e4-e5e7-44fe-9b5d-ab43e21859b2

It has FPGA-inspired reconfigurable fabric!

Bye

P.S.: Shit, I should have such a little toy
compiler as well somewhere in my notes I took
during a lecture. Array with For Loop to

model a hardware bus is really funny.

Mild Shock schrieb:

Hi,

Is there a silver lining of AI democratization?
With MedGamma I can analyse my own broken ribs.
If only I had a body scanner. I am currently exploring

options of LLM models that I could run on
my new AI soaked AMD Ryzen AI 7 350 laptop.
While Qualcomm spear headed LLM players with

their LM Studio. There is FastFlowLM
that can do Ryzen, and would utilize the NPU. For
example to run a distilled DeepSeek would amount to:

flm run deepseek-r1:8b

And yes there is MedGamma:

MedGemma:4B (Multimodal) Running Exclusively on AMD Ryzenrao AI NPU https://www.youtube.com/watch?v=KWzXZEOcgK4

Bye

Mild Shock schrieb:

Hi,

How it started (released January 20, 2025):

Despite trailing the U.S., China continues to make
significant strides in AI development. A notable example
is the Chinese startup DeepSeek, which released an AI
model named R1. This model demonstrates reasoning
capabilities comparable to leading U.S. models and is
available as open-source software, allowing free use and
commercialization. However, R1 faces issues with censorship,
likely influenced by Chinese regulations.
https://en.wikipedia.org/wiki/DeepSeek#Release_history

How its going (released January 21, 2025):

In a strategic move to further solidify its leadership,
President Donald Trump announced the Stargate initiative.
This $500 billion investment, a collaboration between OpenAI,
Oracle, and SoftBank, aims to develop extensive AI
infrastructure across the U.S., including data centers
and energy facilities. The project is expected to create
over 100,000 jobs and enhance the nation's AI capabilities.
https://en.wikipedia.org/wiki/The_Stargate_Project

With SoftBank they basically call Japan for help!

LoL

Bye

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.prolog

Hi,

Was Linus Torvalds cautious or clueless?

"I think AI is really interesting and I think it
is going to change the world. At the same time,
I hate the hype cycle so much that I really don't
want to go there. So, my approach to AI right now
is I will basically ignore it because I think
the whole tech industry around AI is in a
very bad position, and its 90% marketing and
10% reality. And, in 5 years, things will change
and at that point, we will see what of the AI
is getting used for real workloads".

https://www.tweaktown.com/news/101381/linux-creator-linus-torvalds-ai-is-useless-its-90-marketing-while-he-ignores-for-now/index.html

I think his fallacy is to judge AI as hype.
So his timeline 2030 might have received a
suckerpunch by Copilot+ already now in late

2025. Before in 2024, when he made his statement,
AI was already not hype at all:

2009rCo2012 (Deep Learning Wave): GPUs began being
used for deep learning research, thanks to frameworks
like Caffe and Theano. This was when convolutional
networks for vision really took off.

2012rCo2015 (Big Data + Deep Learning): Data centers
started leveraging clusters of GPUs for large-scale
training, using distributed frameworks like
TensorFlow and PyTorch (from 2016). Text analysis
and recommendation systems were already benefiting from this.

2015rCo2020 (Specialized Accelerators): Companies
like Google (TPU), Nvidia (A100), and Qualcomm
(Hexagon DSP) developed purpose-built hardware
for AI inference and training. Large-scale NLP
models like BERT were trained in these environments.

2020rCo2024 (Commercial AI Explosion): On-device AI,
cloud AI services, Copilot+, Claude integrations rCo
all of these are the practical realization of what
had been quietly powering research and enterprise
workloads for over a decade.

Bye


--- Synchronet 3.21a-Linux NewsLink 1.2