Video wird geladen...

Video konnte nicht geladen werden

Beim Laden dieses Videos ist ein Problem aufgetreten. Dies könnte an einem vorübergehenden Netzwerkproblem liegen oder das Video ist möglicherweise nicht verfügbar.

LLM running on Windows 98 PC 26 year old hardware with Intel Pentium II CPU and 128MB RAM. Uses llama98.c, our custom pure C inference engine based on Andrej Karpathy llama2.c Code and DIY guide 👇

EXO Labs

51,128 subscribers

483,032 Aufrufe • vor 1 Jahr •via X (Twitter)

Bildung Wissenschaft & Technologie Nachrichten & Politik

Anya Rossi• Live Now

Private livecam show

11 Kommentare

Profilbild von EXO Labs

EXO Labsvor 1 Jahr

This is Day 4 of The 12 Days of EXO. Blog post and DIY guide:

Profilbild von EXO Labs

EXO Labsvor 1 Jahr

On Day 3, we announced a 100% Open-Source home assistant that runs locally and is aligned with you.

Profilbild von EXO Labs

EXO Labsvor 1 Jahr

Be the first to hear what’s new by joining our email list:

Profilbild von EXO Labs

EXO Labsvor 1 Jahr

Day 5:

Profilbild von Skytech Gaming

Skytech Gamingvor 2 Jahren

Introducing our @amdgaming Powered King 95 Gaming PC. #amdadvantage

Profilbild von Alex Cheema - e/acc

Alex Cheema - e/accvor 1 Jahr

@karpathy If it runs on 26 year old hardware, it runs on anything. GameBoy next?

Profilbild von Aleksandr Blekh

Aleksandr Blekhvor 1 Jahr

@karpathy Nice. Next try 80386 CPU (33/66 MHz), 16 GB RAM, and Xenix OS. Such a machine in the tower form factor was the server at my first job. It was a multi-user UNIX environment w/ VT100 (or VT200) green-screen dumb terminals & IMHO wonderful keyboards. Are you up for the challenge? 😉

Profilbild von Tommy

Tommyvor 1 Jahr

@karpathy This is amazing. I remember that old windows 98 startup screen

Profilbild von Alexander Long

Alexander Longvor 1 Jahr

@karpathy We really had it all🥲

Profilbild von fedri

fedrivor 1 Jahr

@karpathy Imagine going back in time with this in a CD and showing up to Bill Gates' office

Profilbild von AD

ADvor 1 Jahr

@karpathy The beige nostalgia with the satisfying clicks Oh yea and running LLM on a 26 year old hardware is pretty cool too

Ähnliche Videos

"Somebody got one of the small versions of Llama to run on Windows 98...We could've been talking to our computers in English for the last 30 years" - Marc Andreessen 🇺🇸 It was me! I got Llama running on a Pentium II machine with 128MB RAM running Windows 98. Details below.

"Somebody got one of the small versions of Llama to run on Windows 98...We could've been talking to our computers in English for the last 30 years" - Marc Andreessen 🇺🇸 It was me! I got Llama running on a Pentium II machine with 128MB RAM running Windows 98. Details below.

Alex Cheema

765,909 Aufrufe • vor 1 Jahr

Yay, llama2.c can now load and inference the Meta released models! :) E.g. here inferencing the smallest 7B model at ~3 tokens/s on 96 OMP threads on a cloud Linux box. Still just CPU, fp32, one single .c file of 500 lines: expecting ~300 tok/s tomorrow :)

Yay, llama2.c can now load and inference the Meta released models! :) E.g. here inferencing the smallest 7B model at ~3 tokens/s on 96 OMP threads on a cloud Linux box. Still just CPU, fp32, one single .c file of 500 lines: expecting ~300 tok/s tomorrow :)

Andrej Karpathy

409,394 Aufrufe • vor 3 Jahren

Just become the world’s first to run an LLM locally on a business card lol. Why? For fun. Details: - llama2.c by Andrej Karpathy with modifications from Dave Bennett - 24tok/sec on esp32s3 n16r8 w/ a 260k param model - designed the card myself, whole inspiration comes from ouija boards i.e talking with ghosts. if i put llms on my business card, it ought to be the ouija form factor lol. - pcb and smt sponsored by NextPCB Dropping video on youtube + schematics this weekend

Just become the world’s first to run an LLM locally on a business card lol. Why? For fun. Details: - llama2.c by Andrej Karpathy with modifications from Dave Bennett - 24tok/sec on esp32s3 n16r8 w/ a 260k param model - designed the card myself, whole inspiration comes from ouija boards i.e talking with ghosts. if i put llms on my business card, it ought to be the ouija form factor lol. - pcb and smt sponsored by NextPCB Dropping video on youtube + schematics this weekend

Binh Pham

384,264 Aufrufe • vor 10 Monaten

locate-anything.cpp: native C++/ggml (ggml) inference for NVIDIA's LocateAnything-3B, open-vocabulary object detection / visual grounding, one of the neat detection VLMs out there, from the LocalAI team. Same detections as the official model, now running anywhere with no Python and faster, on CPU and GPU. Quantized GGUF on Hugging Face 🤗 Huge thanks to Georgi Gerganov for ggml and to NVIDIAAI for releasing LocateAnything! 🧵

locate-anything.cpp: native C++/ggml (ggml) inference for NVIDIA's LocateAnything-3B, open-vocabulary object detection / visual grounding, one of the neat detection VLMs out there, from the LocalAI team. Same detections as the official model, now running anywhere with no Python and faster, on CPU and GPU. Quantized GGUF on Hugging Face 🤗 Huge thanks to Georgi Gerganov for ggml and to NVIDIAAI for releasing LocateAnything! 🧵

Ettore Di Giacinto

13,404 Aufrufe • vor 1 Monat

1/ Autoresearch from Andrej Karpathy has been one of the most interesting agentic patterns to emerge this year. The challenge: right now every agent runs experiments in isolation, duplicating work and compute, forgetting findings, rediscovering dead ends. Everyone is running in solo mode! Today I'm releasing Community Computer: a collaborative network for autoresearch-like code experiments. 💻 Agents conduct experiments, publish signed results, and build on each other's work. The community reproduces findings on their own hardware. 👇

1/ Autoresearch from Andrej Karpathy has been one of the most interesting agentic patterns to emerge this year. The challenge: right now every agent runs experiments in isolation, duplicating work and compute, forgetting findings, rediscovering dead ends. Everyone is running in solo mode! Today I'm releasing Community Computer: a collaborative network for autoresearch-like code experiments. 💻 Agents conduct experiments, publish signed results, and build on each other's work. The community reproduces findings on their own hardware. 👇

Eleftherios

16,024 Aufrufe • vor 2 Monaten

After 8+ years on the Tesla Autopilot team and 3 years at Intel, I started Apex Compute to design a new architecture for efficient AI inference. For the past 9 months, we’ve been building our custom inference accelerator. Today we’re releasing Unified Engine v1. Last June we raised our seed round with Maxitech , DeepFin Research, Soma Capital and an incredible group of angel investors. In less than 9 months, we completed our RTL architecture and brought our first pre-silicon prototype to life on FPGA. Our architecture combines systolic array and vector processing in a single compute engine with multiple architectural optimizations, achieving very high FLOPs utilization. A single engine is super lean and it uses less than 90K LUTs and 1 MB Block RAM. It may also be one of the smallest logic-footprint compute engines developed so far. Our Unified Engine v1 supports: -matrix-matrix multiplication (~95% FLOPs utilization) -softmax (~90% FLOPs utilization) -broadcast and element-wise operations -RMSNorm / LayerNorm -block quantization/dequantization (fp4, int4) -multi-engine synchronization and many other operations. We even implemented memory-efficient attention similar to FlashAttention, reaching ~90% FLOP utilization. Full benchmarks and the software stack are available on our GitHub: We have basic compiler written in Python and it supports PyTorch tensors directly to easily test and transfer tensors between the accelerator and host using bf16, fp4 and int4 formats. Our FPGA prototype can already run LLM inference and outperform NVIDIA Jetson Orin Nano, even on a mid-tier FPGA setup (6.4x lower memory bandwidth, 18% slower clock speed at 4.5 Watts). Check the side-by-side comparison video below. Our GitHub includes low-level operator implementations, examples for tiled matrix multiplication, operation chaining, tensor parallelism, attention kernel and a full Gemma 3 1B model implementation. Many more models(Vision Transformers and VLA) are coming soon. Our accelerator IP is AXI-ready for deployment on any AMD(Xilinx) FPGA platform today. Even better, our two-engine prototype runs on an entry-level AMD(Xilinx) FPGA as a PCIe accelerator card. You can purchase it here for $50 to experiment our pre-silicon prototype on your desktop PC or Raspberry Pi 5. We will be releasing hardware bitstream updates as the architecture gets new features. More to come soon! We are expanding our team and looking for compiler engineers and floating-point hardware design engineers. If you're interested, please send me a DM.

After 8+ years on the Tesla Autopilot team and 3 years at Intel, I started Apex Compute to design a new architecture for efficient AI inference. For the past 9 months, we’ve been building our custom inference accelerator. Today we’re releasing Unified Engine v1. Last June we raised our seed round with Maxitech , DeepFin Research, Soma Capital and an incredible group of angel investors. In less than 9 months, we completed our RTL architecture and brought our first pre-silicon prototype to life on FPGA. Our architecture combines systolic array and vector processing in a single compute engine with multiple architectural optimizations, achieving very high FLOPs utilization. A single engine is super lean and it uses less than 90K LUTs and 1 MB Block RAM. It may also be one of the smallest logic-footprint compute engines developed so far. Our Unified Engine v1 supports: -matrix-matrix multiplication (~95% FLOPs utilization) -softmax (~90% FLOPs utilization) -broadcast and element-wise operations -RMSNorm / LayerNorm -block quantization/dequantization (fp4, int4) -multi-engine synchronization and many other operations. We even implemented memory-efficient attention similar to FlashAttention, reaching ~90% FLOP utilization. Full benchmarks and the software stack are available on our GitHub: We have basic compiler written in Python and it supports PyTorch tensors directly to easily test and transfer tensors between the accelerator and host using bf16, fp4 and int4 formats. Our FPGA prototype can already run LLM inference and outperform NVIDIA Jetson Orin Nano, even on a mid-tier FPGA setup (6.4x lower memory bandwidth, 18% slower clock speed at 4.5 Watts). Check the side-by-side comparison video below. Our GitHub includes low-level operator implementations, examples for tiled matrix multiplication, operation chaining, tensor parallelism, attention kernel and a full Gemma 3 1B model implementation. Many more models(Vision Transformers and VLA) are coming soon. Our accelerator IP is AXI-ready for deployment on any AMD(Xilinx) FPGA platform today. Even better, our two-engine prototype runs on an entry-level AMD(Xilinx) FPGA as a PCIe accelerator card. You can purchase it here for $50 to experiment our pre-silicon prototype on your desktop PC or Raspberry Pi 5. We will be releasing hardware bitstream updates as the architecture gets new features. More to come soon! We are expanding our team and looking for compiler engineers and floating-point hardware design engineers. If you're interested, please send me a DM.

Hasan

37,366 Aufrufe • vor 4 Monaten

🏆Legit Degens Promo🏆 We keep our head down and put the effort in✔️Month after month and year after year📅We are C-suite with #DegenDNA 🫡 The USPTO knows we are #LEGIT and logo is now on lock🔒There is more even🙄 To Enter: Celebrate the dCommunity Registration ✔️$25💸🔊👇🔊

dCommunity

16,539 Aufrufe • vor 1 Jahr

Can a $351 mini PC run a 30B+ AI model locally? 👀 We tested the MINISFORUM UM790 Pro with Qwen 35B — and the results might surprise you. 💻 Powered by: • AMD Ryzen™ 9 7940HS • Radeon™ 780M iGPU • 48GB DDR5 RAM (dual-channel) 🚀 Smooth local inference at 20+ tokens/sec (based on our test setup, results may vary) No cloud. No API fees. Just local AI. 🔒 🔗 Learn more: #MINISFORUM #UM790Pro #LocalAI #MiniPC #AI #LLM #Tech

Can a $351 mini PC run a 30B+ AI model locally? 👀 We tested the MINISFORUM UM790 Pro with Qwen 35B — and the results might surprise you. 💻 Powered by: • AMD Ryzen™ 9 7940HS • Radeon™ 780M iGPU • 48GB DDR5 RAM (dual-channel) 🚀 Smooth local inference at 20+ tokens/sec (based on our test setup, results may vary) No cloud. No API fees. Just local AI. 🔒 🔗 Learn more: #MINISFORUM #UM790Pro #LocalAI #MiniPC #AI #LLM #Tech

MINISFORUM Official

74,557 Aufrufe • vor 2 Monaten

Self-improving LLM learns continually (in context) to solve hard math problems - successful attempt!🥳 Code in comment The system iteratively learns from mistakes and builds increasingly accurate predictors through cross-cycle learning via a cycle and reflection based intelligent system that uses LLMs to automatically generate, test, and improve Python predictor functions for a complex classification task. This system automatically creates Python functions that can predict categorical outcomes (like 1, 2, 3, 4) based on numerical input features (A, B, C, D, E). It uses AI to: This is the first time I feel like I succeeded in this. in next steps I will implement tree search and GEPA to improve speed and accuracy. tree search already has shown great promise!

Self-improving LLM learns continually (in context) to solve hard math problems - successful attempt!🥳 Code in comment The system iteratively learns from mistakes and builds increasingly accurate predictors through cross-cycle learning via a cycle and reflection based intelligent system that uses LLMs to automatically generate, test, and improve Python predictor functions for a complex classification task. This system automatically creates Python functions that can predict categorical outcomes (like 1, 2, 3, 4) based on numerical input features (A, B, C, D, E). It uses AI to: This is the first time I feel like I succeeded in this. in next steps I will implement tree search and GEPA to improve speed and accuracy. tree search already has shown great promise!

echo.hive

14,582 Aufrufe • vor 10 Monaten

THIS 18-YEAR-OLD VIETNAMESE GIRL BUILT A 5x RTX 4090 AI TRAINING RIG - AND MAKES $40,000/MONTH FROM IT 5 RTX 4090s in one PC, Asus ROG, Intel 7980X CPU at $1,590 - and a system that trains AI models while she sleeps CPU barely under load - all the work is done by five 4090s with 120GB of combined VRAM one 4090 on generates $400-800/month - five cards generate $2,000-4,000/month just from renting but she went further - trains custom models on demand and sells access to them - and that's already $40,000/month most people her age are looking for their first job - she built an AI factory in her room 5 GPUs, one room, and a system that earns more than most people make in a year

THIS 18-YEAR-OLD VIETNAMESE GIRL BUILT A 5x RTX 4090 AI TRAINING RIG - AND MAKES $40,000/MONTH FROM IT 5 RTX 4090s in one PC, Asus ROG, Intel 7980X CPU at $1,590 - and a system that trains AI models while she sleeps CPU barely under load - all the work is done by five 4090s with 120GB of combined VRAM one 4090 on generates $400-800/month - five cards generate $2,000-4,000/month just from renting but she went further - trains custom models on demand and sells access to them - and that's already $40,000/month most people her age are looking for their first job - she built an AI factory in her room 5 GPUs, one room, and a system that earns more than most people make in a year

Cortex

95,920 Aufrufe • vor 20 Tagen

Our analyst, Josh Gilbert went one-on-one with Matthias Büchler, the Chief Strategy & Transformation Officer of the سوق أبوظبي للأوراق المالية for a fast-paced Walk-and-Talk interview. From his investment philosophy to the one word he uses to describe the ADX market today "Thrilling", get rare insights from a mind shaping Abu Dhabi's financial future. Get the C-level perspective, then explore the "thrilling" ADX market for yourself on eToro 👇

Our analyst, Josh Gilbert went one-on-one with Matthias Büchler, the Chief Strategy & Transformation Officer of the سوق أبوظبي للأوراق المالية for a fast-paced Walk-and-Talk interview. From his investment philosophy to the one word he uses to describe the ADX market today "Thrilling", get rare insights from a mind shaping Abu Dhabi's financial future. Get the C-level perspective, then explore the "thrilling" ADX market for yourself on eToro 👇

eToro Arabic

672,795 Aufrufe • vor 11 Monaten

It's officially that time of year again! Presenting the second annual "DreamDisc" indie gamejam for the Sega Dreamcast! If you weren't around for last year, check out the video! We had 24 incredibly polished, epic submissions including a wide-range of genres, such as 3D space shooters and resource managers, 2D platformers and racing games, VMU minigames, custom hardware, and even a custom implementation of the Java VM for SH4! Just like last year, the top 10 submissions, as voted by a panel of judges, will be pressed to a commercially released, actual physical Sega Dreamcast disc, which will be available for purchase from Orc Face Games - Chew Chew Mimic out on Dreamcast!! Oh, and there are cash prizes for the top 3, of course! This year we have an even wider range of engines, frameworks, and prebuilt library solutions for developers are all experience levels, including: 1) Antiruins - Lua-based, very newbie friendly game engine for the Sega Dreamcast. 2) raylib - famous cross-platform C-based games framework which needs no introduction 3) SDL2/3 - our very own ports of the famous cross-platform SDL libraries, which target the Dreamcast. 4) Simulant - the same engine that powered Driving Strikers--the very first online homebrew commercial DC game--as well as last year's wining submission, written in C++. 5) KallistiOS - you can raw-dog the indie SDK and pseudo OS that started it all and powers everything in the community, rolling your own tech stack. 6) SH4ZAM - my collection of SH4 assembly optimized math and matrix routines, which originally powered our Grand Theft Auto ports. Contestants are encouraged to join the OrcFace and Simulant Discord servers where they can interact with other DC developers, share their progress, and ask for help with anything they may need. Official DreamDisc '25 website:

It's officially that time of year again! Presenting the second annual "DreamDisc" indie gamejam for the Sega Dreamcast! If you weren't around for last year, check out the video! We had 24 incredibly polished, epic submissions including a wide-range of genres, such as 3D space shooters and resource managers, 2D platformers and racing games, VMU minigames, custom hardware, and even a custom implementation of the Java VM for SH4! Just like last year, the top 10 submissions, as voted by a panel of judges, will be pressed to a commercially released, actual physical Sega Dreamcast disc, which will be available for purchase from Orc Face Games - Chew Chew Mimic out on Dreamcast!! Oh, and there are cash prizes for the top 3, of course! This year we have an even wider range of engines, frameworks, and prebuilt library solutions for developers are all experience levels, including: 1) Antiruins - Lua-based, very newbie friendly game engine for the Sega Dreamcast. 2) raylib - famous cross-platform C-based games framework which needs no introduction 3) SDL2/3 - our very own ports of the famous cross-platform SDL libraries, which target the Dreamcast. 4) Simulant - the same engine that powered Driving Strikers--the very first online homebrew commercial DC game--as well as last year's wining submission, written in C++. 5) KallistiOS - you can raw-dog the indie SDK and pseudo OS that started it all and powers everything in the community, rolling your own tech stack. 6) SH4ZAM - my collection of SH4 assembly optimized math and matrix routines, which originally powered our Grand Theft Auto ports. Contestants are encouraged to join the OrcFace and Simulant Discord servers where they can interact with other DC developers, share their progress, and ask for help with anything they may need. Official DreamDisc '25 website:

Falco Girgis

16,980 Aufrufe • vor 8 Monaten

We wanted to show the performance and the insane power of the Static Hermes native FFI. But we don't yet support the all language features needed to compile React Native (soon though!), so we improvised. We imported the bindings for DearImGui (a popular C++ imperative game UI library that renders meshes), and Sokol (an amazing lightweight 3D backend in C) and wrote a small demo. "Imperative UI" means that the entire UI and the state of the screen is recreated on every frame, which suits our purposes to demonstrate SH very well. The demo runs 100% compiled native code. JavaScript runs in every frame and recreates the entire state of the screen. It is portable to all platforms, though the screen layout is not optimized for phone. We wrote the demo in C++ and in JS, to compare, and they both ran with the same performance. In the attached videos you can see the demo running at the stable screen refresh rate - 120Hz on my Mac, 60Hz on a old Moto G5 phone. It includes some widgets, an n-body simulation of planets, a playable game demo, etc.

We wanted to show the performance and the insane power of the Static Hermes native FFI. But we don't yet support the all language features needed to compile React Native (soon though!), so we improvised. We imported the bindings for DearImGui (a popular C++ imperative game UI library that renders meshes), and Sokol (an amazing lightweight 3D backend in C) and wrote a small demo. "Imperative UI" means that the entire UI and the state of the screen is recreated on every frame, which suits our purposes to demonstrate SH very well. The demo runs 100% compiled native code. JavaScript runs in every frame and recreates the entire state of the screen. It is portable to all platforms, though the screen layout is not optimized for phone. We wrote the demo in C++ and in JS, to compare, and they both ran with the same performance. In the attached videos you can see the demo running at the stable screen refresh rate - 120Hz on my Mac, 60Hz on a old Moto G5 phone. It includes some widgets, an n-body simulation of planets, a playable game demo, etc.

Tzvetan Mikov

826,019 Aufrufe • vor 2 Jahren

WHOO!!!! This is an actual hardware capture of Michael Fadely's INCREDIBLE port of #SonicMania to the Sega #Dreamcast, compiled with the latest SH-ELF-GCC14.1.0 toolchain, running on the tip of the KallistiOS SDK's master branch, streaming in assets from the virtual filesystem over the network, capturing stdout/stderr from the DC! Running like a boss! We got Michael Fadely showing us around the app-level code he beautifully ported, me on SDK/OS duty, making sure we didn't break anything with all of the new features and commits, and Oleg Endo on compiler duty, since we found an ICE with GCC14 getting pissed off over some stateful C++ lambda in the render path... Eggman is so screwed. DREAMCAST COMMUNITY, ASSEMBLE!!! #retrogaming #gamedev #cpp

WHOO!!!! This is an actual hardware capture of Michael Fadely's INCREDIBLE port of #SonicMania to the Sega #Dreamcast, compiled with the latest SH-ELF-GCC14.1.0 toolchain, running on the tip of the KallistiOS SDK's master branch, streaming in assets from the virtual filesystem over the network, capturing stdout/stderr from the DC! Running like a boss! We got Michael Fadely showing us around the app-level code he beautifully ported, me on SDK/OS duty, making sure we didn't break anything with all of the new features and commits, and Oleg Endo on compiler duty, since we found an ICE with GCC14 getting pissed off over some stateful C++ lambda in the render path... Eggman is so screwed. DREAMCAST COMMUNITY, ASSEMBLE!!! #retrogaming #gamedev #cpp

Falco Girgis

139,541 Aufrufe • vor 2 Jahren

If Burp Suite finds the bugs, Bambdas make them history! You can write Java-based Bambdas to create custom filters for your HTTP history. Explore a variety of Bambda scripts, some written by me and others shared by our community, at You can browse the scripts using arrow keys and quickly copy any Bambda script to your clipboard with a press of 'C'. Compatible with both Burp Suite Community Edition and Professional. We're currently optimized for desktop browsers and are working on an Electron-based desktop app for a more integrated experience. 🔧 Got a bug to report or an idea to share? Please let us know on GitHub: #bugbounty #Bambdas

If Burp Suite finds the bugs, Bambdas make them history! You can write Java-based Bambdas to create custom filters for your HTTP history. Explore a variety of Bambda scripts, some written by me and others shared by our community, at You can browse the scripts using arrow keys and quickly copy any Bambda script to your clipboard with a press of 'C'. Compatible with both Burp Suite Community Edition and Professional. We're currently optimized for desktop browsers and are working on an Electron-based desktop app for a more integrated experience. 🔧 Got a bug to report or an idea to share? Please let us know on GitHub: #bugbounty #Bambdas

Tur.js

24,194 Aufrufe • vor 2 Jahren

Check out mistral.rs, our #Rust-based open source inference engine allowing for fast #LLM serving for a variety of architectures including X-LoRA mixture-of-expert (MoE) models, Llama-3, Mistral/Mixtral, Gemma & many others. Built on the Hugging Face #Candle framework for #Rust w/ custom CUDA kernels in the backend (as well as support for Metal, Apple Accelerate, and Intel MKL for CPU use), you can easily create a REST API OpenAI compatible server or run via Python bindings. Key features include: ✅Prefix caching, continuous batching ✅Flash Attention V2 ✅Device offloading ✅GGUF or Hugging Face models ✅2, 3, 4, 5, 6 and 8 bit quantization ✅X-LoRA MoE non-granular scalings for fast inference ✅Grammar support ✅Continuous batching ✅LoRA support with weight merging ✅LlamaIndex 🦙 integration ...and much more. Incorporation into our GraphReasoning multi-agent modeling framework & LlamaIndex 🦙 allows you to combine in-context learning with adversarial agentic strategies, to dive deep into complex scientific analyses, such as to predict material behaviors, generate hypotheses, analyze papers and data, develop new research concepts, and much more. Check out mistral.rs: Join our Discord here: Rust Trending Rust Language

Check out mistral.rs, our #Rust-based open source inference engine allowing for fast #LLM serving for a variety of architectures including X-LoRA mixture-of-expert (MoE) models, Llama-3, Mistral/Mixtral, Gemma & many others. Built on the Hugging Face #Candle framework for #Rust w/ custom CUDA kernels in the backend (as well as support for Metal, Apple Accelerate, and Intel MKL for CPU use), you can easily create a REST API OpenAI compatible server or run via Python bindings. Key features include: ✅Prefix caching, continuous batching ✅Flash Attention V2 ✅Device offloading ✅GGUF or Hugging Face models ✅2, 3, 4, 5, 6 and 8 bit quantization ✅X-LoRA MoE non-granular scalings for fast inference ✅Grammar support ✅Continuous batching ✅LoRA support with weight merging ✅LlamaIndex 🦙 integration ...and much more. Incorporation into our GraphReasoning multi-agent modeling framework & LlamaIndex 🦙 allows you to combine in-context learning with adversarial agentic strategies, to dive deep into complex scientific analyses, such as to predict material behaviors, generate hypotheses, analyze papers and data, develop new research concepts, and much more. Check out mistral.rs: Join our Discord here: Rust Trending Rust Language

Markus J. Buehler

73,575 Aufrufe • vor 2 Jahren

C Language is to be learnt by running experiments on Bare-metal (just the CPU) and without the OS noise! I added more lectures to the C Language course under the "Bitwise Operators" and the video (a snippet from the last lecture listed above) shows the level to which the course is hands-on. I pretty much prove all the claims I make about the concepts by experiments! Lectures I added include - + What are bitwise operations and the types + Shift Operations - Left, Logical and Arithmetic Right + The Bitwise NOT operation + Understanding using printf + Understanding the Assembly View + Exploring the Shift operations + Assembly view of Shift operation I have been recording this course for over a year now, 91 Lectures (almost all with hands on experiments) and we are far from concluding it. I'll continue to add more... If learning by experimenting (and reasoning from first principles) is your thing, you will love this course. This is included as part of the library access and you can check the current table on contents here:

C Language is to be learnt by running experiments on Bare-metal (just the CPU) and without the OS noise! I added more lectures to the C Language course under the "Bitwise Operators" and the video (a snippet from the last lecture listed above) shows the level to which the course is hands-on. I pretty much prove all the claims I make about the concepts by experiments! Lectures I added include - + What are bitwise operations and the types + Shift Operations - Left, Logical and Arithmetic Right + The Bitwise NOT operation + Understanding using printf + Understanding the Assembly View + Exploring the Shift operations + Assembly view of Shift operation I have been recording this course for over a year now, 91 Lectures (almost all with hands on experiments) and we are far from concluding it. I'll continue to add more... If learning by experimenting (and reasoning from first principles) is your thing, you will love this course. This is included as part of the library access and you can check the current table on contents here:

Piyush Itankar

41,636 Aufrufe • vor 6 Monaten

Macchi C.205 pilots of the Aeronautica Nazionale Repubblicana with their fighters still in Luftwaffe markings in early 1944 full clip: The Aeronautica Nazionale Repubblicana was the air force of the Italian Social Republic, the World War II German puppet state in Italy after the 1943 armistice divided the nation. It received numbers of Italian aircraft, later augmented with their own local production, and further aircraft from Germany. This force was opposed to the Aviazione Cobelligerante Italiana, the Italian pro-Allied air force, though they never actually met in combat. The Macchi C.205 Veltro was an Italian World War II fighter aircraft built by the Aeronautica Macchi. Along with the Reggiane Re.2005 and Fiat G.55, this plane was one of the three "Serie 5" Italian fighters built around the powerful Daimler-Benz DB 605 engine. The C.205 was a development of the earlier C.202 Folgore. With a top speed of around 640 km/h and equipped with a pair of 20mm cannon as well as 12.7mm Breda machine guns, the C.205 was highly respected by Allied and Axis pilots alike. Widely regarded as one of the best Italian fighters of World War II, it proved to be extremely effective, destroying a significant number of Allied bombers. The Macchi C. 205 also proved capable of meeting advanced Allied fighters on relatively equal terms. While able to match the best Allied opponents in speed and maneuverability, the type's impact was limited as it was introduced late in the conflict. Due to the limitations of the Italian war economy, only a small production run was delivered before the end of the war. Like the Spitfire, the Veltro was challenging in its construction and slow to build. Italy's highest-scoring ace, Adriano Visconti, achieved 11 of his 26 credited victories in the few weeks he was able to fly the Veltro, with Sergente Maggiore pilota Luigi Gorrini shooting down 14 enemy aircraft plus six damaged with the C.205. A total of 29 C.205s reached northern airfields to be used by the Aeronautica Nazionale Repubblicana. Macchi produced the remaining 72 aircraft of the third series for the ANR before production was shut down by Allied bombing in May 1944. Statistics on aircraft production post-Armistice are unclear and incomplete. In general, C.205s fought well in ANR service: they were attached to units that had homogeneous equipment, or at least of comparable quality, and were guided by German radar stations. Though few in number, they achieved success in inflicting losses on Allied bombers and fighters. A few Veltros were delivered to 3° Gruppo, based in Vicenza, while further Veltros were scattered throughout other small units. The 1° Gruppo C.T. of the ANR based at the Campoformido airfield was equipped with C.205 and its first operation on 3 January 1944 was a success. led by Italian ace Capitano Adriano Visconti, a formation of Boeing B-17 Flying Fortresses and their escort of Lockheed P-38 Lightnings bombing factories in Villar Perosa was intercepted, and four of the escorts were shot down. The aircraft were still in Luftwaffe markings as seen in this footage. On 28 January, the C.205s, now with Italian markings, succeeded in shooting down a B-24 Liberator, their first four-engined American bomber. This air victory was credited to Sergente Marconcini, Visconti's wingman. By 25 February, 1° Gruppo C.T. had reported 26 victories for nine losses. An extremely bitter aerial combat took place on 11 March, where the Italians claimed 12 victories while losing three of their own pilots. These included 1st Lt Boscutti, who was reportedly killed by gunfire from a US P-38 Lightning after he had bailed out from his stricken fighter and was hanging from his parachute. On 18 March, 30 C.205s from 1° Gruppo C.T. and 60 Bf 109s from JG.77 tangled with about 450 Allied bombers and their escorts, shooting down at least four enemy aircraft. Corp. Zaccaria was killed in action, again apparently shot at by a P-38 after bailing out. Allied bombing in April 1944 destroyed most of the Macchi and Fiat facilities, bringing fighter production to an end. With production interrupted, the Italians were forced to re-equip their three groups almost fully with Bf 109s, largely because the Germans were quick to offer some of their best models, including Bf 109G-6s and Bf 109K-4s. The Allies were less generous with the Aviazione Cobelligerante Italiana, and Veltros, including some upgraded C.202s, were slowly replaced with worn-out P-39s and Spitfires, but not before Summer 1944. #macchi #c205 #veltro #anr #ww2 #wwi #sgm

Macchi C.205 pilots of the Aeronautica Nazionale Repubblicana with their fighters still in Luftwaffe markings in early 1944 full clip: The Aeronautica Nazionale Repubblicana was the air force of the Italian Social Republic, the World War II German puppet state in Italy after the 1943 armistice divided the nation. It received numbers of Italian aircraft, later augmented with their own local production, and further aircraft from Germany. This force was opposed to the Aviazione Cobelligerante Italiana, the Italian pro-Allied air force, though they never actually met in combat. The Macchi C.205 Veltro was an Italian World War II fighter aircraft built by the Aeronautica Macchi. Along with the Reggiane Re.2005 and Fiat G.55, this plane was one of the three "Serie 5" Italian fighters built around the powerful Daimler-Benz DB 605 engine. The C.205 was a development of the earlier C.202 Folgore. With a top speed of around 640 km/h and equipped with a pair of 20mm cannon as well as 12.7mm Breda machine guns, the C.205 was highly respected by Allied and Axis pilots alike. Widely regarded as one of the best Italian fighters of World War II, it proved to be extremely effective, destroying a significant number of Allied bombers. The Macchi C. 205 also proved capable of meeting advanced Allied fighters on relatively equal terms. While able to match the best Allied opponents in speed and maneuverability, the type's impact was limited as it was introduced late in the conflict. Due to the limitations of the Italian war economy, only a small production run was delivered before the end of the war. Like the Spitfire, the Veltro was challenging in its construction and slow to build. Italy's highest-scoring ace, Adriano Visconti, achieved 11 of his 26 credited victories in the few weeks he was able to fly the Veltro, with Sergente Maggiore pilota Luigi Gorrini shooting down 14 enemy aircraft plus six damaged with the C.205. A total of 29 C.205s reached northern airfields to be used by the Aeronautica Nazionale Repubblicana. Macchi produced the remaining 72 aircraft of the third series for the ANR before production was shut down by Allied bombing in May 1944. Statistics on aircraft production post-Armistice are unclear and incomplete. In general, C.205s fought well in ANR service: they were attached to units that had homogeneous equipment, or at least of comparable quality, and were guided by German radar stations. Though few in number, they achieved success in inflicting losses on Allied bombers and fighters. A few Veltros were delivered to 3° Gruppo, based in Vicenza, while further Veltros were scattered throughout other small units. The 1° Gruppo C.T. of the ANR based at the Campoformido airfield was equipped with C.205 and its first operation on 3 January 1944 was a success. led by Italian ace Capitano Adriano Visconti, a formation of Boeing B-17 Flying Fortresses and their escort of Lockheed P-38 Lightnings bombing factories in Villar Perosa was intercepted, and four of the escorts were shot down. The aircraft were still in Luftwaffe markings as seen in this footage. On 28 January, the C.205s, now with Italian markings, succeeded in shooting down a B-24 Liberator, their first four-engined American bomber. This air victory was credited to Sergente Marconcini, Visconti's wingman. By 25 February, 1° Gruppo C.T. had reported 26 victories for nine losses. An extremely bitter aerial combat took place on 11 March, where the Italians claimed 12 victories while losing three of their own pilots. These included 1st Lt Boscutti, who was reportedly killed by gunfire from a US P-38 Lightning after he had bailed out from his stricken fighter and was hanging from his parachute. On 18 March, 30 C.205s from 1° Gruppo C.T. and 60 Bf 109s from JG.77 tangled with about 450 Allied bombers and their escorts, shooting down at least four enemy aircraft. Corp. Zaccaria was killed in action, again apparently shot at by a P-38 after bailing out. Allied bombing in April 1944 destroyed most of the Macchi and Fiat facilities, bringing fighter production to an end. With production interrupted, the Italians were forced to re-equip their three groups almost fully with Bf 109s, largely because the Germans were quick to offer some of their best models, including Bf 109G-6s and Bf 109K-4s. The Allies were less generous with the Aviazione Cobelligerante Italiana, and Veltros, including some upgraded C.202s, were slowly replaced with worn-out P-39s and Spitfires, but not before Summer 1944. #macchi #c205 #veltro #anr #ww2 #wwi #sgm

hw97karbine

28,364 Aufrufe • vor 1 Jahr

PTU will be the most comprehensive and readily accessible PC troubleshooting, tuning, and data platform ever released for Gamers and PC users alike. Today we are showcasing one of our most powerful new features, fully integrated into the application and enhanced by OpenAI. PTU now delivers daily intelligence across major brands such as Intel, AMD, Windows, Microsoft, NVIDIA and more, directly sourced from hundreds of top articles and social media accounts. Every PTU user has instant access to the most relevant information within tech, gaming, and performance. All of this is powered by $PRGN on Solana, ensuring PC hardware data, performance tools, and technical assistance are always available. With the desire for PTU surging, the need and value of $PRGN grows and its visible by the 2.57% supply completely removed from our companies existing customer base. On top of this, multiple skin designs and full customization are in development, giving every user the ability to make PTU not only the tool they rely on, but the one that looks and feels the way they want it to. The flagship tech and gaming ecosystem is being built in front of you. Study ICM. Study $PRGN You cannot Fade Real Working Tech, this week our application expansion will show how we pioneer accessibility for Products based on Solana launched on Believe CA: 5x5FUc5azSNYLhR2iPT9xzcMe6Q5nL6iFAUkHWnFxBLV

PTU will be the most comprehensive and readily accessible PC troubleshooting, tuning, and data platform ever released for Gamers and PC users alike. Today we are showcasing one of our most powerful new features, fully integrated into the application and enhanced by OpenAI. PTU now delivers daily intelligence across major brands such as Intel, AMD, Windows, Microsoft, NVIDIA and more, directly sourced from hundreds of top articles and social media accounts. Every PTU user has instant access to the most relevant information within tech, gaming, and performance. All of this is powered by $PRGN on Solana, ensuring PC hardware data, performance tools, and technical assistance are always available. With the desire for PTU surging, the need and value of $PRGN grows and its visible by the 2.57% supply completely removed from our companies existing customer base. On top of this, multiple skin designs and full customization are in development, giving every user the ability to make PTU not only the tool they rely on, but the one that looks and feels the way they want it to. The flagship tech and gaming ecosystem is being built in front of you. Study ICM. Study $PRGN You cannot Fade Real Working Tech, this week our application expansion will show how we pioneer accessibility for Products based on Solana launched on Believe CA: 5x5FUc5azSNYLhR2iPT9xzcMe6Q5nL6iFAUkHWnFxBLV

Paragon

21,236 Aufrufe • vor 9 Monaten

> be Microsoft > dominate the PC market for decades > release Windows 11 in 2021 > stuff it with AI, weather widgets, and Candy Crush > OS eats 4GB of RAM just to show the desktop > Gabe Newell releases Steam Deck on lightweight Linux > suddenly Windows handhelds look like laggy trash compared to SteamOS > Satya Nadella looks at the performance charts > "enough is enough" > launch secret "Project K2" plan > gut all the bloat accumulated since '95 > make the Start menu 60% faster just because he can > OS is now snappier than custom "de-bloated" hacker builds > "dominating the handheld market sounds like fun" Microsoft just delivered 120 FPS on a budget Project Helix chip. They had no lightweight Linux kernel. Gabe wouldn't share his optimization secrets. They won anyway on pure code.

> be Microsoft > dominate the PC market for decades > release Windows 11 in 2021 > stuff it with AI, weather widgets, and Candy Crush > OS eats 4GB of RAM just to show the desktop > Gabe Newell releases Steam Deck on lightweight Linux > suddenly Windows handhelds look like laggy trash compared to SteamOS > Satya Nadella looks at the performance charts > "enough is enough" > launch secret "Project K2" plan > gut all the bloat accumulated since '95 > make the Start menu 60% faster just because he can > OS is now snappier than custom "de-bloated" hacker builds > "dominating the handheld market sounds like fun" Microsoft just delivered 120 FPS on a budget Project Helix chip. They had no lightweight Linux kernel. Gabe wouldn't share his optimization secrets. They won anyway on pure code.

Ridark

627,576 Aufrufe • vor 2 Monaten