Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > sci.logic > #348207 > unrolled thread

The Wuhan Virus that destroyed Python [ggml Manifesto]

Back to article view | Back to sci.logic

Contents

  The Wuhan Virus that destroyed Python [ggml Manifesto] Mild Shock <janburse@fastmail.fm> - 2026-07-22 21:01 +0200
    Deadlock Exorcism: Switch from Push to Pull [A pi-calculus Specification of Prolog] (Re: The Wuhan Virus that destroyed Python [ggml Manifesto]) Mild Shock <janburse@fastmail.fm> - 2026-07-23 00:25 +0200
      Why do you even need a mpmc queue? [Thunder Kittens] (Re: Deadlock Exorcism: Switch from Push to Pull) Mild Shock <janburse@fastmail.fm> - 2026-07-23 08:45 +0200
        Trivial balancing example for (int i=0; i<global_id; i++) (Re: Why do you even need a mpmc queue? [Thunder Kittens]) Mild Shock <janburse@fastmail.fm> - 2026-07-23 08:55 +0200
          The Pixel Phone AI Experiment Song (Enqueue/dequeue need not be fast and can spinn ["fairness" questions]) Mild Shock <janburse@fastmail.fm> - 2026-07-23 09:19 +0200
            Enqueue/dequeue need not be fast and can spinn ["fairness" questions] (Re: The Pixel Phone AI Experiment Song (Enqueue/dequeue need not be fast and can spinn ["fairness" questions]) Mild Shock <janburse@fastmail.fm> - 2026-07-23 09:23 +0200
        And, where did I talk about rockets? [Hint its about xAI's Grok] (Re: Why do you even need a mpmc queue? [Thunder Kittens]) Mild Shock <janburse@fastmail.fm> - 2026-07-25 01:25 +0200
          Why forget something, that was never on my mind (Re: And, where did I talk about rockets? [Hint its about xAI's Grok]) Mild Shock <janburse@fastmail.fm> - 2026-07-25 09:49 +0200
        Example Mandel Brot rendering [Faster with MIMD] (Was: Why do you even need a mpmc queue? [Thunder Kittens]) Mild Shock <janburse@fastmail.fm> - 2026-07-25 09:56 +0200
    Potential Python Recovery: Free Threading [3.13 release] (Re: The Wuhan Virus that destroyed Python [ggml Manifesto]) Mild Shock <janburse@fastmail.fm> - 2026-07-23 10:21 +0200
    The things XILINX braught to the AMD table (Re: The Wuhan Virus that destroyed Python [ggml Manifesto]) Mild Shock <janburse@fastmail.fm> - 2026-07-23 18:48 +0200
      NVIDIA evacuated its Chinese market [Tau Scaling] (Re: The things XILINX braught to the AMD table) Mild Shock <janburse@fastmail.fm> - 2026-07-23 19:13 +0200
        Micro penis mother sung arias (Re: NVIDIA evacuated its Chinese market [Tau Scaling]) Mild Shock <janburse@fastmail.fm> - 2026-07-24 14:40 +0200
          Micro penis brain is in constant hiatus (Re: Micro penis mother sung arias) Mild Shock <janburse@fastmail.fm> - 2026-07-24 15:27 +0200
            Ignoramus or Ignorabimus: I don't care [(Re: Micro penis brain is in constant hiatus (Re: Micro penis mother sung arias) Mild Shock <janburse@fastmail.fm> - 2026-07-24 15:35 +0200
              You are a moron, brainless putin payed (Re: Ignoramus or Ignorabimus: I don't care) Mild Shock <janburse@fastmail.fm> - 2026-07-24 18:01 +0200
                Yeah keep reading my posts, uninspired fool (Re: You are a moron, brainless putin payed) Mild Shock <janburse@fastmail.fm> - 2026-07-24 19:47 +0200
              Out of the blue accusation span 15 days [Empirical USENET study] (Re: Ignoramus or Ignorabimus: I don't care) Mild Shock <janburse@fastmail.fm> - 2026-07-24 20:27 +0200
              A brain desease of 20 days [Rossy Boy] (Re: Ignoramus or Ignorabimus: I don't care) Mild Shock <janburse@fastmail.fm> - 2026-07-29 18:41 +0200
                I didn't use a Ryzen Halo, whats wrong with you? (Re: A brain desease of 20 days [Rossy Boy]) Mild Shock <janburse@fastmail.fm> - 2026-07-29 23:25 +0200
                Ignoramus / Ignorabimus Barometer: Almost 1 Month (Re: A brain desease of 20 days [Rossy Boy]) Mild Shock <janburse@fastmail.fm> - 2026-08-03 00:07 +0200
        Re: NVIDIA evacuated its Chinese market [Tau Scaling] (Re: The things XILINX braught to the AMD table) Mild Shock <janburse@fastmail.fm> - 2026-07-28 14:17 +0200
        ASML stocks are plunging, bye bye dutchies (Re: NVIDIA evacuated its Chinese market [Tau Scaling]) Mild Shock <janburse@fastmail.fm> - 2026-07-28 14:18 +0200
    Little Data Center on Your Palm [AI Laptops for 500 USD] (Re: The Wuhan Virus that destroyed Python [ggml Manifesto]) Mild Shock <janburse@fastmail.fm> - 2026-07-24 17:59 +0200
      2008: 4 Blades + Tesla S1070 versus 2026: 1 AI Laptop (Re: Little Data Center on Your Palm [AI Laptops for 500 USD]) Mild Shock <janburse@fastmail.fm> - 2026-07-24 18:15 +0200
      Budget AI Laptop 2026 versus Cray T3D 1995 (Was: Little Data Center on Your Palm) Mild Shock <janburse@fastmail.fm> - 2026-08-05 14:21 +0200
        Re: Budget AI Laptop 2026 versus Cray T3D 1995 R Kym Horsell <kym@sdf.org> - 2026-08-05 21:11 +0000
          AI Alarmist with Supercomputer on Yacht [Horsy Boy] (Was: Budget AI Laptop 2026 versus Cray T3D 1995) Mild Shock <janburse@fastmail.fm> - 2026-08-06 13:21 +0200
            Re: AI Alarmist with Supercomputer on Yacht [Horsy Boy] R Kym Horsell <kym@sdf.org> - 2026-08-06 11:56 +0000
    Hurry the blue bus doesnt stop indefinitely (Re: The Wuhan Virus that destroyed Python [ggml Manifesto]) Mild Shock <janburse@fastmail.fm> - 2026-07-24 20:37 +0200
      Not SIMD, a MIMD design for NVIDIA Volta (Re: Hurry the blue bus doesnt stop indefinitely) Mild Shock <janburse@fastmail.fm> - 2026-07-24 20:58 +0200
        Could take 3-4 months find machine / browser (Re: Not SIMD, a MIMD design for NVIDIA Volta) Mild Shock <janburse@fastmail.fm> - 2026-07-24 21:16 +0200
        The Koan of pi-WAM queues [FORTRAN-S] (Re: Not SIMD, a MIMD design for NVIDIA Volta) Mild Shock <janburse@fastmail.fm> - 2026-07-26 19:54 +0200
          The turbo capping of AI Laptops (Was: The Koan of pi-WAM queues [FORTRAN-S]) Mild Shock <janburse@fastmail.fm> - 2026-07-26 20:00 +0200
          Re: The Koan of pi-WAM queues [FORTRAN-S] (Re: Not SIMD, a MIMD design for NVIDIA Volta) Mild Shock <janburse@fastmail.fm> - 2026-07-27 09:16 +0200
          Why forget Bulgarians, never on my mind (Re: The Koan of pi-WAM queues [FORTRAN-S]) Mild Shock <janburse@fastmail.fm> - 2026-07-27 09:16 +0200
            miniTriton CUDA is an alternative to torch variants (Re: Why forget Bulgarians, never on my mind) Mild Shock <janburse@fastmail.fm> - 2026-07-27 09:52 +0200
              Andrej Karpathy original gangster of Budget Laptop (Re: miniTriton CUDA is an alternative to torch variants) Mild Shock <janburse@fastmail.fm> - 2026-07-27 09:54 +0200
            The evolution of hardware and GPT-2 training (Re: Why forget Bulgarians, never on my mind) Mild Shock <janburse@fastmail.fm> - 2026-07-27 10:57 +0200
              How speed up π-WAM with vector operations (Re: The evolution of hardware and GPT-2 training) Mild Shock <janburse@fastmail.fm> - 2026-07-27 11:10 +0200
                AI accelerator extend from GPU to CPU [Zero Copying] (Re: How speed up π-WAM with vector operations) Mild Shock <janburse@fastmail.fm> - 2026-07-27 13:21 +0200
                  The invention of vector and matrix registers [NVIDIA Volta] (Re: AI accelerator extend from GPU to CPU [Zero Copying]) Mild Shock <janburse@fastmail.fm> - 2026-07-27 13:22 +0200
                    Re: The invention of vector and matrix registers [NVIDIA Volta] (Re: AI accelerator extend from GPU to CPU [Zero Copying]) Ross Finlayson <ross.a.finlayson@gmail.com> - 2026-07-27 07:34 -0700
                      Maybe they should have named it NVIDIA Einstein [Rossy Boy Toe Sucking] (Was: The invention of vector and matrix registers [NVIDIA Volta]) Mild Shock <janburse@fastmail.fm> - 2026-07-27 17:12 +0200
                      Re: The invention of vector and matrix registers [NVIDIA Volta] (Re: AI accelerator extend from GPU to CPU [Zero Copying]) R Kym Horsell <kym@sdf.com> - 2026-07-27 15:43 +0000
                        Re: The invention of vector and matrix registers [NVIDIA Volta] (Re: AI accelerator extend from GPU to CPU [Zero Copying]) R Kym Horsell <kymhorsell@gmail.com> - 2026-07-27 15:46 +0000
                        π-WAM is not adding decimals, it is removing decimals (Was: The invention of vector and matrix registers [NVIDIA Volta]) Mild Shock <janburse@fastmail.fm> - 2026-07-27 18:34 +0200
                          In Budget Laptops the TOPS come with low energy footprint (Re: π-WAM is not adding decimals, it is removing decimals) Mild Shock <janburse@fastmail.fm> - 2026-07-27 18:45 +0200
      Potato Computer owner impressed by Ukraine Tech [Rossy Boys Brother?] (Was: Hurry the blue bus doesnt stop indefinitely) Mild Shock <janburse@fastmail.fm> - 2026-07-27 16:56 +0200
        Rossy Boy is neither Einstein nor Zweistein (Was: Potato Computer owner impressed by Ukraine Tech) Mild Shock <janburse@fastmail.fm> - 2026-07-27 18:25 +0200
          You are still chewing on SIMD. LoL (Re: Rossy Boy is neither Einstein nor Zweistein) Mild Shock <janburse@fastmail.fm> - 2026-07-29 17:12 +0200
            Hurry Rossy Boy, the blue bus is waiting (Re: You are still chewing on SIMD. LoL) Mild Shock <janburse@fastmail.fm> - 2026-07-29 17:53 +0200
              Look how they advertized CUDA and logical threads (Re: Hurry Rossy Boy, the blue bus is waiting) Mild Shock <janburse@fastmail.fm> - 2026-07-29 17:55 +0200
                Forget any arithmetization of product FSA (Re: Look how they advertized CUDA and logical threads) Mild Shock <janburse@fastmail.fm> - 2026-07-29 17:58 +0200
            Rossy Boys tears could cool a data center [pi-WAM Interleaved Synchronized Emulator] (Re: You are still chewing on SIMD. LoL) Mild Shock <janburse@fastmail.fm> - 2026-07-29 20:06 +0200
              I don't use Rust, you are crazy [Jump off a bridge, idiot] (Re: Rossy Boys tears could cool a data center [pi-WAM Interleaved Synchronized Emulator]) Mild Shock <janburse@fastmail.fm> - 2026-07-29 20:25 +0200
                Hack ecosystem ignorance paired with paranoia [Nand to Tetris] (Re: I don't use Rust, you are crazy) Mild Shock <janburse@fastmail.fm> - 2026-07-29 22:52 +0200
                  A funny Q16.16 experiment with Hack (Re: Hack ecosystem ignorance paired with paranoia [Nand to Tetris]) Mild Shock <janburse@fastmail.fm> - 2026-07-29 23:11 +0200
                    Summer Challenge: libSQL = Prolog+Modes [VDBE versus π-WAM] (Was: A funny Q16.16 experiment with Hack) Mild Shock <janburse@fastmail.fm> - 2026-07-30 11:25 +0200
                  I wrote Hack VM for π-WAM from scratch [4 Months total JavaScript, Python and Java] (Re: Hack ecosystem ignorance paired with paranoia) Mild Shock <janburse@fastmail.fm> - 2026-07-30 19:37 +0200
                    For WebGPU I first had SIMD in mind (Re: I wrote Hack VM for π-WAM from scratch) Mild Shock <janburse@fastmail.fm> - 2026-07-30 19:51 +0200
                      Corr.: 4 Months --> 4 Weeks (Re: For WebGPU I first had SIMD in mind) Mild Shock <janburse@fastmail.fm> - 2026-07-30 20:05 +0200
                    MIPS is a big Huffman mess [But Hack could do it] (Re: I wrote Hack VM for π-WAM from scratch) Mild Shock <janburse@fastmail.fm> - 2026-07-30 22:32 +0200
                      Not declarative with PHI (Φ) nodes (Was: MIPS is a big Huffman mess [But Hack could do it]) Mild Shock <janburse@fastmail.fm> - 2026-07-30 22:38 +0200
                      Not declarative with PHI (Φ) nodes (Re: MIPS is a big Huffman mess [But Hack could do it]) Mild Shock <janburse@fastmail.fm> - 2026-07-30 22:39 +0200
                      Re: MIPS is a big Huffman mess [But Hack could do it] (Re: I wrote Hack VM for π-WAM from scratch) Ross Finlayson <ross.a.finlayson@gmail.com> - 2026-07-31 09:52 -0700
                        Re: MIPS is a big Huffman mess [But Hack could do it] (Re: I wrote Hack VM for π-WAM from scratch) Ross Finlayson <ross.a.finlayson@gmail.com> - 2026-07-31 10:00 -0700
                          Re: MIPS is a big Huffman mess [But Hack could do it] (Re: I wrote Hack VM for π-WAM from scratch) Ross Finlayson <ross.a.finlayson@gmail.com> - 2026-07-31 10:07 -0700
                    Quo Vadis: Extend investigations to WebNN (Was: I wrote Hack VM for π-WAM from scratch [4 Months total JavaScript, Python and Java]) Mild Shock <janburse@fastmail.fm> - 2026-07-31 20:43 +0200
                Standing on the shoulders of giants (Re: I don't use Rust, you are crazy [Jump off a bridge, idiot]) Mild Shock <janburse@fastmail.fm> - 2026-08-04 03:19 +0200
                  You Thief! Stealing Szemeredi, Aristotle, Leibniz, etc.. (Re: Standing on the shoulders of giants) Mild Shock <janburse@fastmail.fm> - 2026-08-04 15:19 +0200
                    How Rossy Boys plagiarism works [Copy Paste Slop] (Re: You Thief! Stealing Szemeredi, Aristotle, Leibniz, etc.. ) Mild Shock <janburse@fastmail.fm> - 2026-08-04 17:58 +0200
                      Statistics gave up, no salient truth [Signal Collapse] Re: How Rossy Boys plagiarism works [Copy Paste Slop] (Re: You Thief! Stealing Szemeredi, Aristotle, Leibniz, etc.. ) Mild Shock <janburse@fastmail.fm> - 2026-08-04 18:17 +0200
      The Paul Armer Square Revisited (Was: "Mathematics in the Age of AI") Mild Shock <janburse@fastmail.fm> - 2026-08-19 23:49 +0200
        Communism will Save Us! [Pivot Russia for China] (Re: The Paul Armer Square Revisited) Mild Shock <janburse@fastmail.fm> - 2026-08-20 13:15 +0200
          How to increase your "Convincingness" [Anthropic AI Text Hacked] Mild Shock <janburse@fastmail.fm> - 2026-08-20 19:31 +0200
            Free Speech for (my) Robots [Kristallnaach BAP] (Re: How to increase your "Convincingness") Mild Shock <janburse@fastmail.fm> - 2026-08-20 19:49 +0200
              Reality of Proof Assistants / Coding [Luhmanns Zettelkasten] (Re: Free Speech for (my) Robots) Mild Shock <janburse@fastmail.fm> - 2026-08-20 20:23 +0200
    Got it. Or are you too stupid? [New Usenet Mantra] (Was: The Wuhan Virus that destroyed Python [ggml Manifesto]) Mild Shock <janburse@fastmail.fm> - 2026-07-27 18:59 +0200
    Lamas in a cradle and Lamas on the edge [Red Pyjama] (Re: The Wuhan Virus that destroyed Python [ggml Manifesto]) Mild Shock <janburse@fastmail.fm> - 2026-07-29 13:03 +0200
      AI Accelerators and ISO Prolog multi-threading (Re: Lamas in a cradle and Lamas on the edge [Red Pyjama]) Mild Shock <janburse@fastmail.fm> - 2026-07-29 17:02 +0200
        Actor/Erlang is dead, no Thread and Mailbox conflation [golang channels] (Re: AI Accelerators and ISO Prolog multi-threading) (Re: Lamas in a cradle and Lamas on the edge [Red Pyjama]) Mild Shock <janburse@fastmail.fm> - 2026-07-29 17:04 +0200
          Can library(ironpaw) repurpose FFT hardware [Glimps into Ryzen AI 7 350] (Re: Actor/Erlang is dead, no Thread and Mailbox conflation ) Mild Shock <janburse@fastmail.fm> - 2026-08-01 02:32 +0200
      Tablet and phone UBS-C remote debugging (Re: Lamas in a cradle and Lamas on the edge [Red Pyjama]) Mild Shock <janburse@fastmail.fm> - 2026-08-01 12:19 +0200
        NPUs doing 2d chess comms (Manhattan Distance or L1 Norm) (Re: Tablet and phone UBS-C remote debugging) Mild Shock <janburse@fastmail.fm> - 2026-08-01 14:12 +0200
          NACK retransmission might double Manhattan Distance (Re: NPUs doing 2d chess comms) Mild Shock <janburse@fastmail.fm> - 2026-08-01 14:24 +0200
            I am using WebGPU, and not WebGL (Re: NACK retransmission might double Manhattan Distance) Mild Shock <janburse@fastmail.fm> - 2026-08-02 00:47 +0200
              Texture inside my compute shader makes no sense (Re: I am using WebGPU, and not WebGL) Mild Shock <janburse@fastmail.fm> - 2026-08-02 02:40 +0200
                Prolog inferencing and not canvasing fancy stuff (Re: Texture inside my compute shader makes no sense) Mild Shock <janburse@fastmail.fm> - 2026-08-02 02:42 +0200
                  It’s called . . . . enshittification (About the price tag for using a multifile/1) Mild Shock <janburse@fastmail.fm> - 2026-08-14 00:49 +0200
        Chris M. Thomasson can ask 100 more questions (Was: Tablet and phone UBS-C remote debugging) Mild Shock <janburse@fastmail.fm> - 2026-08-02 02:46 +0200
          npm install webgpu [Google Dawn] (Re: Chris M. Thomasson can ask 100 more questions) Mild Shock <janburse@fastmail.fm> - 2026-08-02 03:03 +0200
            GPU elasticity was already invented in 2008 with CUDA (Re: npm install webgpu [Google Dawn]) Mild Shock <janburse@fastmail.fm> - 2026-08-03 00:01 +0200
      Synthetic Multilanguage Autoformalization Dataset [Informath project] (Was: Lamas in a cradle and Lamas on the edge [Red Pyjama]) Mild Shock <janburse@fastmail.fm> - 2026-08-08 09:17 +0200
        Re: Synthetic Multilanguage Autoformalization Dataset [Informath project] (Was: Lamas in a cradle and Lamas on the edge [Red Pyjama]) x3 <x@x.net> - 2026-08-08 11:45 -0700
          Nice try Rossy Boy --> **plonk** (Was: Synthetic Multilanguage Autoformalization Dataset [Informath project]) Mild Shock <janburse@fastmail.fm> - 2026-08-08 23:02 +0200
            Ethernal September Idiots Gone (Was: Nice try Rossy Boy --> **plonk**) Mild Shock <janburse@fastmail.fm> - 2026-08-08 23:15 +0200
        Six Proofs and Generally Inteligent Systems [EyeProlog Pseudo Scientism] (Re: Synthetic Multilanguage Autoformalization Dataset [Informath project]) Mild Shock <janburse@fastmail.fm> - 2026-08-15 15:19 +0200
          Everybody does eat and sleep [The SK hynix Story] (Was: Six Proofs and Generally Intelligent Systems [EyeProlog Pseudo Scientism]) Mild Shock <janburse@fastmail.fm> - 2026-08-15 18:47 +0200
    Even send_color and recv_color can block [Cerebras Waver] (Re: The Wuhan Virus that destroyed Python [ggml Manifesto]) Mild Shock <janburse@fastmail.fm> - 2026-08-02 23:39 +0200
    GPU Elasticity: Collective Communications Libraries (Re: The Wuhan Virus that destroyed Python [ggml Manifesto]) Mild Shock <janburse@fastmail.fm> - 2026-08-07 14:35 +0200
      What are Flits and Phits? [Network on a Chip] (Was: GPU Elasticity: Collective Communications Libraries) Mild Shock <janburse@fastmail.fm> - 2026-08-07 18:05 +0200
        The Mac Neo is a Budget Monster [GPU Channels] (Re: What are Flits and Phits? [Network on a Chip]) Mild Shock <janburse@fastmail.fm> - 2026-08-11 16:25 +0200
          The luminaries of duct-tape engineering [Sweeney and Torvald] (Re: The Mac Neo is a Budget Monster [GPU Channels]) Mild Shock <janburse@fastmail.fm> - 2026-08-11 16:54 +0200
    Cristallina: Thank you for the Beam (Was: The Wuhan Virus that destroyed Python [ggml Manifesto]) Mild Shock <janburse@fastmail.fm> - 2026-08-09 21:19 +0200
      Re: Cristallina: Thank you for the Beam (Was: The Wuhan Virus that destroyed Python [ggml Manifesto]) Mild Shock <janburse@fastmail.fm> - 2026-08-18 14:56 +0200
      Giant Shoulders for AI Surprises (Was: Cristallina: Thank you for the Beam) Mild Shock <janburse@fastmail.fm> - 2026-08-18 14:57 +0200
        Loderunner Enemy AI better than SWI-Prolog? [Prolog Education Group] Mild Shock <janburse@fastmail.fm> - 2026-08-18 16:23 +0200
          How to shoot yourself in the foot (Was: Loderunner Enemy AI better than SWI-Prolog? [Prolog Education Group]) Mild Shock <janburse@fastmail.fm> - 2026-08-18 16:42 +0200
          William A. Howard solved all his 99 problems (Re: Loderunner Enemy AI better than SWI-Prolog?) Mild Shock <janburse@fastmail.fm> - 2026-08-23 01:47 +0200

Page 6 of 6 — ← Prev page 1 2 3 4 5 [6]

#348335 — GPU Elasticity: Collective Communications Libraries (Re: The Wuhan Virus that destroyed Python [ggml Manifesto])

Hi,

How it started, NVIDIA being cool:

NCCL provides routines such as all-gather,
all-reduce, broadcast, reduce, reduce-scatter,
and point-to-point send and receive. These
routines are optimized to achieve high
bandwidth and low latency over PCIe,
NVIDIA NVLink™, and other high-speed
interconnects within a node and over
NVIDIA networking across nodes.
https://developer.nvidia.com/nccl

How its going, vLLM trying to be cool:

[RFC]: Native Weight Syncing APIs
However, there are no standardized methods for
performing online weight syncing. Open source projects
like SkyRL, VeRL, and TRL need to include their
own implementations of the weight syncing
infrastructure, leading to added complexity
for developers seeking to adopt vLLM as their
inference server for post-training workloads.
https://github.com/vllm-project/vllm/issues/31848

How much Workers are enough? I guess it depends
on I/O parallelism, CPU Memory parallelism, CPU
Processing parallelism, and now also

GPU Memory parallelism and GPU Processing
parallelism, and last but least you might have
a couple DMAs sitting here and there,

or even invoking a sort of RDMA. Quite amazing!

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 “Okay, 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?”
> 
> Why I was able to do it so quickly - basically,
> for all that I saw it’s pretty much GPT-J architecture
> with some modifications, like some extra memorization
> layers. It’s 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

[toc] | [prev] | [next] | [standalone]

#348336 — What are Flits and Phits? [Network on a Chip] (Was: GPU Elasticity: Collective Communications Libraries)

Hi,

Recently there was a paper somebody mentioning
a flit doing a ACK or NACK, to express
backpressure inside a Network on a Chip.

But what is a flit? It seems multiple
flits can be used to create the message
passing in one directiob before the

ACK or NACK in the other direction?

"The growing need for performance from
computing systems drove the industry into
the multi-core and many-core arena. In this
setup, the execution of a kernel (a program)
is split across multiple processors and the
computation happens in parallel

Flits represent logical units of information,
while phits represent the physical domain,
that is, phits represent the number of bits
that can be transferred in parallel in a
single cycle. Consider the Cray T3D. It has
an interconnection network which uses

flit level message flow control wherein each
flit is composed of eight 16-bit phits. That
means its flit size is 128bits and phit size
is 16bits. Also consider the IBM SP2 switch.
It also uses the flit level message flow
control, but its flit size is equal to its
phit size, which is set to 8 bits."
https://en.wikipedia.org/wiki/Flit_(computer_networking)#Example

Well my idea how this is realized in silicon
is rather foggy, I mean even the Hack project
from Nand 2 Tetris, does not show some gate level
schemes for flits and phits.

Could be an interesting extension. But somehow
the image of flits and phits inspired my channel
objects here below. But I am afraid they are fire
and forget, no ACK and NACK:

π-WAM Contest: 1 Million Packets with Prolog
https://medium.com/2989/ec3e91551773

Its amazing that a max_size(1) buffer
can beat an unbounded buffer!

LoL

Bye

Mild Shock schrieb:
> Hi,
> 
> How it started, NVIDIA being cool:
> 
> NCCL provides routines such as all-gather,
> all-reduce, broadcast, reduce, reduce-scatter,
> and point-to-point send and receive. These
> routines are optimized to achieve high
> bandwidth and low latency over PCIe,
> NVIDIA NVLink™, and other high-speed
> interconnects within a node and over
> NVIDIA networking across nodes.
> https://developer.nvidia.com/nccl
> 
> How its going, vLLM trying to be cool:
> 
> [RFC]: Native Weight Syncing APIs
> However, there are no standardized methods for
> performing online weight syncing. Open source projects
> like SkyRL, VeRL, and TRL need to include their
> own implementations of the weight syncing
> infrastructure, leading to added complexity
> for developers seeking to adopt vLLM as their
> inference server for post-training workloads.
> https://github.com/vllm-project/vllm/issues/31848
> 
> How much Workers are enough? I guess it depends
> on I/O parallelism, CPU Memory parallelism, CPU
> Processing parallelism, and now also
> 
> GPU Memory parallelism and GPU Processing
> parallelism, and last but least you might have
> a couple DMAs sitting here and there,
> 
> or even invoking a sort of RDMA. Quite amazing!
> 
> 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 “Okay, 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?”
>>
>> Why I was able to do it so quickly - basically,
>> for all that I saw it’s pretty much GPT-J architecture
>> with some modifications, like some extra memorization
>> layers. It’s 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
> 

[toc] | [prev] | [next] | [standalone]

#348356 — The Mac Neo is a Budget Monster [GPU Channels] (Re: What are Flits and Phits? [Network on a Chip])

Hi,

Now I implemented some multiple producer
and multiple consumer channel objects for
WebGPU. The only API to integrate it user

facing into pi-WAM is this single predicate:

/**
  * flit(C):
  * The predicate succeeds in C with a new channel. The channel
  * can be used from within GPU backed π-WAM logical threads.
  */

The Mac Neo is a Budget Monster. While the
Ryzen AI Laptop cost around 1300.- CHF.
The Mac Neo was around 600.- CHF with all

extras. Here some performance results,
checking out whether channel objects scale,
when increasing their number to

communicate the same 1 millon packets:

Java performance:

AI Laptop    Single    Double
Ryzen    705.1    337.4
Neo    669.4    239.9

WebGPU performance:

AI Laptop    Single    Double
Ryzen    731.8    392.9
Neo    932.8    483.5

Cool! Java is also pretty cool, their
semaphore library is top notch. I couldn't
replicate the resulst with JavaScript yet,

seems their Atomics.wait() resp. Atomics.waitAsync()
is totally broken, using futex is mutex for
fools somehow. I also found some gremlins

attacking one of the GPUs. The Intel AI Laptop
fails the above experiment. Maybe its a driver
Vulkan versus OpenCL or something problem,

or the Lunar lake architecture is nonsense.

Bye

Mild Shock schrieb:
> Hi,
> 
> Recently there was a paper somebody mentioning
> a flit doing a ACK or NACK, to express
> backpressure inside a Network on a Chip.
> 
> But what is a flit? It seems multiple
> flits can be used to create the message
> passing in one directiob before the
> 
> ACK or NACK in the other direction?
> 
> "The growing need for performance from
> computing systems drove the industry into
> the multi-core and many-core arena. In this
> setup, the execution of a kernel (a program)
> is split across multiple processors and the
> computation happens in parallel
> 
> Flits represent logical units of information,
> while phits represent the physical domain,
> that is, phits represent the number of bits
> that can be transferred in parallel in a
> single cycle. Consider the Cray T3D. It has
> an interconnection network which uses
> 
> flit level message flow control wherein each
> flit is composed of eight 16-bit phits. That
> means its flit size is 128bits and phit size
> is 16bits. Also consider the IBM SP2 switch.
> It also uses the flit level message flow
> control, but its flit size is equal to its
> phit size, which is set to 8 bits."
> https://en.wikipedia.org/wiki/Flit_(computer_networking)#Example
> 
> Well my idea how this is realized in silicon
> is rather foggy, I mean even the Hack project
> from Nand 2 Tetris, does not show some gate level
> schemes for flits and phits.
> 
> Could be an interesting extension. But somehow
> the image of flits and phits inspired my channel
> objects here below. But I am afraid they are fire
> and forget, no ACK and NACK:
> 
> π-WAM Contest: 1 Million Packets with Prolog
> https://medium.com/2989/ec3e91551773
> 
> Its amazing that a max_size(1) buffer
> can beat an unbounded buffer!
> 
> LoL
> 
> Bye
> 
> Mild Shock schrieb:
>> Hi,
>>
>> How it started, NVIDIA being cool:
>>
>> NCCL provides routines such as all-gather,
>> all-reduce, broadcast, reduce, reduce-scatter,
>> and point-to-point send and receive. These
>> routines are optimized to achieve high
>> bandwidth and low latency over PCIe,
>> NVIDIA NVLink™, and other high-speed
>> interconnects within a node and over
>> NVIDIA networking across nodes.
>> https://developer.nvidia.com/nccl
>>
>> How its going, vLLM trying to be cool:
>>
>> [RFC]: Native Weight Syncing APIs
>> However, there are no standardized methods for
>> performing online weight syncing. Open source projects
>> like SkyRL, VeRL, and TRL need to include their
>> own implementations of the weight syncing
>> infrastructure, leading to added complexity
>> for developers seeking to adopt vLLM as their
>> inference server for post-training workloads.
>> https://github.com/vllm-project/vllm/issues/31848
>>
>> How much Workers are enough? I guess it depends
>> on I/O parallelism, CPU Memory parallelism, CPU
>> Processing parallelism, and now also
>>
>> GPU Memory parallelism and GPU Processing
>> parallelism, and last but least you might have
>> a couple DMAs sitting here and there,
>>
>> or even invoking a sort of RDMA. Quite amazing!
>>
>> 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 “Okay, 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?”
>>>
>>> Why I was able to do it so quickly - basically,
>>> for all that I saw it’s pretty much GPT-J architecture
>>> with some modifications, like some extra memorization
>>> layers. It’s 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
>>
> 

[toc] | [prev] | [next] | [standalone]

#348357 — The luminaries of duct-tape engineering [Sweeney and Torvald] (Re: The Mac Neo is a Budget Monster [GPU Channels])

Hi,

We didn't find yet a library for our Think that
would support webgpu on the ARM architecture,
so its back to the browser flag and testing there.

The node.js package comes only with:

dist
+-- d3dcompiler_47.dll
+-- darwin-universal.dawn.node
+-- linux-arm64.dawn.node
+-- linux-x64.dawn.node
+-- win32-x64.dawn.node

Its a similar situation like with SVN. In some
communities its not common to provide a ARM build.
They rather use x86 till the end of the universe.

Although we think initiatives like the x86 Ecosystem
Advisory Group could be a clever marketing trick to
hide a funeral service. Adding "luminaries" such as

Tim Sweeney and Linus Torvald to the panel, is even
more so a joke, given that intel produces mutex bottlenecks
instead of futex, where f stands for fast, in their GPU

infrastructure. So who is the teacher and who are
the students? But why even try to create a collation
against ARM, it doesn't make any sense.

Bye

Mild Shock schrieb:
> Hi,
> 
> Now I implemented some multiple producer
> and multiple consumer channel objects for
> WebGPU. The only API to integrate it user
> 
> facing into pi-WAM is this single predicate:
> 
> /**
>   * flit(C):
>   * The predicate succeeds in C with a new channel. The channel
>   * can be used from within GPU backed π-WAM logical threads.
>   */
> 
> The Mac Neo is a Budget Monster. While the
> Ryzen AI Laptop cost around 1300.- CHF.
> The Mac Neo was around 600.- CHF with all
> 
> extras. Here some performance results,
> checking out whether channel objects scale,
> when increasing their number to
> 
> communicate the same 1 millon packets:
> 
> Java performance:
> 
> AI Laptop    Single    Double
> Ryzen    705.1    337.4
> Neo    669.4    239.9
> 
> WebGPU performance:
> 
> AI Laptop    Single    Double
> Ryzen    731.8    392.9
> Neo    932.8    483.5
> 
> Cool! Java is also pretty cool, their
> semaphore library is top notch. I couldn't
> replicate the resulst with JavaScript yet,
> 
> seems their Atomics.wait() resp. Atomics.waitAsync()
> is totally broken, using futex is mutex for
> fools somehow. I also found some gremlins
> 
> attacking one of the GPUs. The Intel AI Laptop
> fails the above experiment. Maybe its a driver
> Vulkan versus OpenCL or something problem,
> 
> or the Lunar lake architecture is nonsense.
> 
> Bye
> 
> Mild Shock schrieb:
>> Hi,
>>
>> Recently there was a paper somebody mentioning
>> a flit doing a ACK or NACK, to express
>> backpressure inside a Network on a Chip.
>>
>> But what is a flit? It seems multiple
>> flits can be used to create the message
>> passing in one directiob before the
>>
>> ACK or NACK in the other direction?
>>
>> "The growing need for performance from
>> computing systems drove the industry into
>> the multi-core and many-core arena. In this
>> setup, the execution of a kernel (a program)
>> is split across multiple processors and the
>> computation happens in parallel
>>
>> Flits represent logical units of information,
>> while phits represent the physical domain,
>> that is, phits represent the number of bits
>> that can be transferred in parallel in a
>> single cycle. Consider the Cray T3D. It has
>> an interconnection network which uses
>>
>> flit level message flow control wherein each
>> flit is composed of eight 16-bit phits. That
>> means its flit size is 128bits and phit size
>> is 16bits. Also consider the IBM SP2 switch.
>> It also uses the flit level message flow
>> control, but its flit size is equal to its
>> phit size, which is set to 8 bits."
>> https://en.wikipedia.org/wiki/Flit_(computer_networking)#Example
>>
>> Well my idea how this is realized in silicon
>> is rather foggy, I mean even the Hack project
>> from Nand 2 Tetris, does not show some gate level
>> schemes for flits and phits.
>>
>> Could be an interesting extension. But somehow
>> the image of flits and phits inspired my channel
>> objects here below. But I am afraid they are fire
>> and forget, no ACK and NACK:
>>
>> π-WAM Contest: 1 Million Packets with Prolog
>> https://medium.com/2989/ec3e91551773
>>
>> Its amazing that a max_size(1) buffer
>> can beat an unbounded buffer!
>>
>> LoL
>>
>> Bye
>>
>> Mild Shock schrieb:
>>> Hi,
>>>
>>> How it started, NVIDIA being cool:
>>>
>>> NCCL provides routines such as all-gather,
>>> all-reduce, broadcast, reduce, reduce-scatter,
>>> and point-to-point send and receive. These
>>> routines are optimized to achieve high
>>> bandwidth and low latency over PCIe,
>>> NVIDIA NVLink™, and other high-speed
>>> interconnects within a node and over
>>> NVIDIA networking across nodes.
>>> https://developer.nvidia.com/nccl
>>>
>>> How its going, vLLM trying to be cool:
>>>
>>> [RFC]: Native Weight Syncing APIs
>>> However, there are no standardized methods for
>>> performing online weight syncing. Open source projects
>>> like SkyRL, VeRL, and TRL need to include their
>>> own implementations of the weight syncing
>>> infrastructure, leading to added complexity
>>> for developers seeking to adopt vLLM as their
>>> inference server for post-training workloads.
>>> https://github.com/vllm-project/vllm/issues/31848
>>>
>>> How much Workers are enough? I guess it depends
>>> on I/O parallelism, CPU Memory parallelism, CPU
>>> Processing parallelism, and now also
>>>
>>> GPU Memory parallelism and GPU Processing
>>> parallelism, and last but least you might have
>>> a couple DMAs sitting here and there,
>>>
>>> or even invoking a sort of RDMA. Quite amazing!
>>>
>>> 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 “Okay, 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?”
>>>>
>>>> Why I was able to do it so quickly - basically,
>>>> for all that I saw it’s pretty much GPT-J architecture
>>>> with some modifications, like some extra memorization
>>>> layers. It’s 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
>>>
>>
> 

[toc] | [prev] | [next] | [standalone]

#348345 — Cristallina: Thank you for the Beam (Was: The Wuhan Virus that destroyed Python [ggml Manifesto])

Hi,

How it started:

Filming a vitamin B12 photoreceptor in action
https://www.psi.ch/de/news/science-features/filming-a-vitamin-b12-photoreceptor-in-action

How its going:

Elon Musk's potential FEL route could challenge EUV lithography
https://www.kucoin.com/news/flash/elon-musk-s-potential-fel-route-could-challenge-euv-lithography

Who will win the Nano Atom mover race,

will the USA OutChip its competitor China
and its supplier Asia in the next years?

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 “Okay, 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?”
> 
> Why I was able to do it so quickly - basically,
> for all that I saw it’s pretty much GPT-J architecture
> with some modifications, like some extra memorization
> layers. It’s 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

[toc] | [prev] | [next] | [standalone]

#348385 — Re: Cristallina: Thank you for the Beam (Was: The Wuhan Virus that destroyed Python [ggml Manifesto])

Hi,

Obviously to generate AI Surprises,
you have to stand on the shoulder of
giants. Otherwise you will not discover

surprises, right? Only mediocre deja vues.
For this purpose, and maybe otherwise
related, in a discussion concerning the future

of the library, Marvin Minsky and Edward A.
Feigenbaum endorsed the idea for books
to ‘talk to each other’:

LET DOCUMENTS TALK TO EACH OTHER
Z. CHEN - 1993
doi.org/10.1108/eb026910

Now we have:

THE MACHINES ARE STUDYING
THE HUMANS ARE SCROLLING
https://9gag.com/gag/a9yQ16o

Bye

P.S.: But who was Marvin Minsky, and would it
be important to use symbolic AI?

The Perceptron Controversy
Yuxi Liu - 2024
https://yuxi.ml/essays/perceptron-controversy

Mild Shock schrieb:
> Hi,
> 
> How it started:
> 
> Filming a vitamin B12 photoreceptor in action
> https://www.psi.ch/de/news/science-features/filming-a-vitamin-b12-photoreceptor-in-action 
> 
> 
> How its going:
> 
> Elon Musk's potential FEL route could challenge EUV lithography
> https://www.kucoin.com/news/flash/elon-musk-s-potential-fel-route-could-challenge-euv-lithography 
> 
> 
> Who will win the Nano Atom mover race,
> 
> will the USA OutChip its competitor China
> and its supplier Asia in the next years?
> 
> 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 “Okay, 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?”
>>
>> Why I was able to do it so quickly - basically,
>> for all that I saw it’s pretty much GPT-J architecture
>> with some modifications, like some extra memorization
>> layers. It’s 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
> 

[toc] | [prev] | [next] | [standalone]

#348386 — Giant Shoulders for AI Surprises (Was: Cristallina: Thank you for the Beam)

Hi,

Obviously to generate AI Surprises,
you have to stand on the shoulder of
giants. Otherwise you will not discover

surprises, right? Only mediocre deja vues.
For this purpose, and maybe otherwise
related, in a discussion concerning the future

of the library, Marvin Minsky and Edward A.
Feigenbaum endorsed the idea for books
to ‘talk to each other’:

LET DOCUMENTS TALK TO EACH OTHER
Z. CHEN - 1993
doi.org/10.1108/eb026910

Now we have:

THE MACHINES ARE STUDYING
THE HUMANS ARE SCROLLING
https://9gag.com/gag/a9yQ16o

Bye

P.S.: But who was Marvin Minsky, and would it
be important to use symbolic AI?

The Perceptron Controversy
Yuxi Liu - 2024
https://yuxi.ml/essays/perceptron-controversy

Mild Shock schrieb:
> Hi,
> 
> How it started:
> 
> Filming a vitamin B12 photoreceptor in action
> https://www.psi.ch/de/news/science-features/filming-a-vitamin-b12-photoreceptor-in-action 
> 
> 
> How its going:
> 
> Elon Musk's potential FEL route could challenge EUV lithography
> https://www.kucoin.com/news/flash/elon-musk-s-potential-fel-route-could-challenge-euv-lithography 
> 
> 
> Who will win the Nano Atom mover race,
> 
> will the USA OutChip its competitor China
> and its supplier Asia in the next years?
> 
> 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 “Okay, 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?”
>>
>> Why I was able to do it so quickly - basically,
>> for all that I saw it’s pretty much GPT-J architecture
>> with some modifications, like some extra memorization
>> layers. It’s 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
> 

[toc] | [prev] | [next] | [standalone]

#348387 — Loderunner Enemy AI better than SWI-Prolog? [Prolog Education Group]

Hi,

Sometimes I think the Loderunner Enemy AI
was way ahead of its time:

Lode Runner - Broderbund - 1983 - Apple II
https://www.youtube.com/watch?v=pJZJepU8law

Meanwhile SWI-Prolog Prologers even don't know
whether a Prolog text is CNF or DNF:

Sets of rules as conjunctions and/or disjunctions
https://swi-prolog.discourse.group/t/sets-of-rules-as-conjunctions-and-or-disjunctions/9779

No Wonder that the EyeProlog "why" feature produces
nonsense. Even a Flea circus is less crazy.

Bye

P.S.: If only there would exist something like
universities where one can take a basic course in
FOL, and then a world wide web, where one can

lookup clark equational theory, clark completion,
curry howard correspondence, etc.. etc..

Mild Shock schrieb:
> Hi,
> 
> Obviously to generate AI Surprises,
> you have to stand on the shoulder of
> giants. Otherwise you will not discover
> 
> surprises, right? Only mediocre deja vues.
> For this purpose, and maybe otherwise
> related, in a discussion concerning the future
> 
> of the library, Marvin Minsky and Edward A.
> Feigenbaum endorsed the idea for books
> to ‘talk to each other’:
> 
> LET DOCUMENTS TALK TO EACH OTHER
> Z. CHEN - 1993
> doi.org/10.1108/eb026910
> 
> Now we have:
> 
> THE MACHINES ARE STUDYING
> THE HUMANS ARE SCROLLING
> https://9gag.com/gag/a9yQ16o
> 
> Bye
> 
> P.S.: But who was Marvin Minsky, and would it
> be important to use symbolic AI?
> 
> The Perceptron Controversy
> Yuxi Liu - 2024
> https://yuxi.ml/essays/perceptron-controversy
> 
> Mild Shock schrieb:
>> Hi,
>>
>> How it started:
>>
>> Filming a vitamin B12 photoreceptor in action
>> https://www.psi.ch/de/news/science-features/filming-a-vitamin-b12-photoreceptor-in-action 
>>
>>
>> How its going:
>>
>> Elon Musk's potential FEL route could challenge EUV lithography
>> https://www.kucoin.com/news/flash/elon-musk-s-potential-fel-route-could-challenge-euv-lithography 
>>
>>
>> Who will win the Nano Atom mover race,
>>
>> will the USA OutChip its competitor China
>> and its supplier Asia in the next years?
>>
>> 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 “Okay, 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?”
>>>
>>> Why I was able to do it so quickly - basically,
>>> for all that I saw it’s pretty much GPT-J architecture
>>> with some modifications, like some extra memorization
>>> layers. It’s 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
>>
> 

[toc] | [prev] | [next] | [standalone]

#348388 — How to shoot yourself in the foot (Was: Loderunner Enemy AI better than SWI-Prolog? [Prolog Education Group])

Hi,

The biggest joke, is to have logic somewhere
in procedure and somewhere in declarative, and
try to see different takes on logic.

They really don't know what this mantra means:

Algorithm = Logic + Control

And cannot relate it to the ideas of declarative
reading and procedural reading. The problem is
a too lax introduction of this notions,

prematurely before the notion "logic" is
even understood. But nobody understands the
meaning of the term "logic", i.e. a set of

supposedly tautological sentences, as under-
stood by mathematical logic. There is a subtle
error again to conflate it with a "calculus",

so this website has a very misleanding title,
although they try hard to not commit the fallacy,
and have subtitles "Proof System" and "Logic Level":

Welcome to LogicProof
https://6ximik9.github.io/naturalDeduction/

The there is this moron:

There is only one “minimal logic”. The term denotes Johansson’s 
Minimalkalkül [1937] — intuitionistic logic without ex falso quodlibet — 
and nothing else. No rival system competes for the name.
https://vidal-rosset.net/rule-correspondence-F.html

Of course there are rival "Proof Systems" aka
calculi, even when the "Logic Level" is minimal
logic. It is as if Joseph Vidal-Rosset doesn't

understand basic German. You have to look behind
"Minimalkalkül" to find "Minimallogic". Right?
To identify calculus with logic, is maybe a 1930's

fallacy, but then we had model theory besides
proof theory, and people should be more educated
now. Model theory can be also expanded to

non-classical logics and even minimal logic, to
give a purely semantic reading. Ok some modern
morons think they need to invoke the word "algebraic".

Bye

Mild Shock schrieb:
> Hi,
> 
> Sometimes I think the Loderunner Enemy AI
> was way ahead of its time:
> 
> Lode Runner - Broderbund - 1983 - Apple II
> https://www.youtube.com/watch?v=pJZJepU8law
> 
> Meanwhile SWI-Prolog Prologers even don't know
> whether a Prolog text is CNF or DNF:
> 
> Sets of rules as conjunctions and/or disjunctions
> https://swi-prolog.discourse.group/t/sets-of-rules-as-conjunctions-and-or-disjunctions/9779 
> 
> 
> No Wonder that the EyeProlog "why" feature produces
> nonsense. Even a Flea circus is less crazy.
> 
> Bye
> 
> P.S.: If only there would exist something like
> universities where one can take a basic course in
> FOL, and then a world wide web, where one can
> 
> lookup clark equational theory, clark completion,
> curry howard correspondence, etc.. etc..
> 
> Mild Shock schrieb:
>> Hi,
>>
>> Obviously to generate AI Surprises,
>> you have to stand on the shoulder of
>> giants. Otherwise you will not discover
>>
>> surprises, right? Only mediocre deja vues.
>> For this purpose, and maybe otherwise
>> related, in a discussion concerning the future
>>
>> of the library, Marvin Minsky and Edward A.
>> Feigenbaum endorsed the idea for books
>> to ‘talk to each other’:
>>
>> LET DOCUMENTS TALK TO EACH OTHER
>> Z. CHEN - 1993
>> doi.org/10.1108/eb026910
>>
>> Now we have:
>>
>> THE MACHINES ARE STUDYING
>> THE HUMANS ARE SCROLLING
>> https://9gag.com/gag/a9yQ16o
>>
>> Bye
>>
>> P.S.: But who was Marvin Minsky, and would it
>> be important to use symbolic AI?
>>
>> The Perceptron Controversy
>> Yuxi Liu - 2024
>> https://yuxi.ml/essays/perceptron-controversy
>>
>> Mild Shock schrieb:
>>> Hi,
>>>
>>> How it started:
>>>
>>> Filming a vitamin B12 photoreceptor in action
>>> https://www.psi.ch/de/news/science-features/filming-a-vitamin-b12-photoreceptor-in-action 
>>>
>>>
>>> How its going:
>>>
>>> Elon Musk's potential FEL route could challenge EUV lithography
>>> https://www.kucoin.com/news/flash/elon-musk-s-potential-fel-route-could-challenge-euv-lithography 
>>>
>>>
>>> Who will win the Nano Atom mover race,
>>>
>>> will the USA OutChip its competitor China
>>> and its supplier Asia in the next years?
>>>
>>> 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 “Okay, 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?”
>>>>
>>>> Why I was able to do it so quickly - basically,
>>>> for all that I saw it’s pretty much GPT-J architecture
>>>> with some modifications, like some extra memorization
>>>> layers. It’s 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
>>>
>>
> 

[toc] | [prev] | [next] | [standalone]

#348401 — William A. Howard solved all his 99 problems (Re: Loderunner Enemy AI better than SWI-Prolog?)

Hi,

Woa! I am impressed:

Howard died in Chicago on March 13, 2026,
at the age of 99.
https://en.wikipedia.org/wiki/William_Alvin_Howard

I recomend every body reading his original Curry-Howard
Isomorphism paper. He even mentions sigma types,
calling it Strong existence (choice operators):

THE FORMULAE-AS-TYPES NOTION OF CONSTRUCTION
https://www.cs.cmu.edu/~crary/819-f09/Howard80.pdf

Can you derive a Prolog program for proof
normalization? A good setting to read the paper
is a greek FKK beach.

Bye

Mild Shock schrieb:
> Hi,
> 
> Sometimes I think the Loderunner Enemy AI
> was way ahead of its time:
> 
> Lode Runner - Broderbund - 1983 - Apple II
> https://www.youtube.com/watch?v=pJZJepU8law
> 
> Meanwhile SWI-Prolog Prologers even don't know
> whether a Prolog text is CNF or DNF:
> 
> Sets of rules as conjunctions and/or disjunctions
> https://swi-prolog.discourse.group/t/sets-of-rules-as-conjunctions-and-or-disjunctions/9779 
> 
> 
> No Wonder that the EyeProlog "why" feature produces
> nonsense. Even a Flea circus is less crazy.
> 
> Bye
> 
> P.S.: If only there would exist something like
> universities where one can take a basic course in
> FOL, and then a world wide web, where one can
> 
> lookup clark equational theory, clark completion,
> curry howard correspondence, etc.. etc..
> 
> Mild Shock schrieb:
>> Hi,
>>
>> Obviously to generate AI Surprises,
>> you have to stand on the shoulder of
>> giants. Otherwise you will not discover
>>
>> surprises, right? Only mediocre deja vues.
>> For this purpose, and maybe otherwise
>> related, in a discussion concerning the future
>>
>> of the library, Marvin Minsky and Edward A.
>> Feigenbaum endorsed the idea for books
>> to ‘talk to each other’:
>>
>> LET DOCUMENTS TALK TO EACH OTHER
>> Z. CHEN - 1993
>> doi.org/10.1108/eb026910
>>
>> Now we have:
>>
>> THE MACHINES ARE STUDYING
>> THE HUMANS ARE SCROLLING
>> https://9gag.com/gag/a9yQ16o
>>
>> Bye
>>
>> P.S.: But who was Marvin Minsky, and would it
>> be important to use symbolic AI?
>>
>> The Perceptron Controversy
>> Yuxi Liu - 2024
>> https://yuxi.ml/essays/perceptron-controversy
>>
>> Mild Shock schrieb:
>>> Hi,
>>>
>>> How it started:
>>>
>>> Filming a vitamin B12 photoreceptor in action
>>> https://www.psi.ch/de/news/science-features/filming-a-vitamin-b12-photoreceptor-in-action 
>>>
>>>
>>> How its going:
>>>
>>> Elon Musk's potential FEL route could challenge EUV lithography
>>> https://www.kucoin.com/news/flash/elon-musk-s-potential-fel-route-could-challenge-euv-lithography 
>>>
>>>
>>> Who will win the Nano Atom mover race,
>>>
>>> will the USA OutChip its competitor China
>>> and its supplier Asia in the next years?
>>>
>>> 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 “Okay, 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?”
>>>>
>>>> Why I was able to do it so quickly - basically,
>>>> for all that I saw it’s pretty much GPT-J architecture
>>>> with some modifications, like some extra memorization
>>>> layers. It’s 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
>>>
>>
> 

[toc] | [prev] | [standalone]

Page 6 of 6 — ← Prev page 1 2 3 4 5 [6]

Back to top | Article view | sci.logic

csiph-web