magic! Open source players successfully used an $8 ESP32-S3 chip to run a 28.9 million parameter large language model offline
Do you believe that the IoT chip ESP32-S3 microcontroller, which is only cheaper than two cups of coffee at US$8 ...
Tag: Github
Microsoft open source Mage-Flow image model: only 4B parameters can generate 1024×1024 images in one second, and the benchmark test is tied with FLUX.2 32B
Microsoft Asia publicly released Mage-Flow on July 22 at Hugging Face
Microsoft VibeVoice-ASR: an open source speech recognition model that transcribes 60 minutes of audio files in one go and provides three-in-one structured output
Microsoft Research recently open sourced VibeVoice, a 7 billion-parameter speech AI model that can transcribe up to 60 words in ...
Draw a robotic arm in one sentence: the open source tool text-to-cad turns Claude Code and Codex into a CAD engineer
In the past, it was almost impossible for an amateur who had never learned any CAD software to create a ...
KTransformers open source framework allows 24GB consumer-grade graphics card to successfully run 671B MoE large model
Recently, China’s Tsinghua University released a set of open source framework GitHub projects called KTransformers, achieving the goal of using ...
The miracle of a 22-year-old handheld console! Foreign gods successfully ran “Absolute Force” on PSP and ran a smooth 60 FPS
Most people probably kept the PSP handheld machine 22 years ago and did not play it. However, the foreign genius ...
OptimizerDuck is an open-source Windows system tuning and acceleration tool with a Traditional Chinese interface.
After using Windows for a while, you may experience slower performance and reduced speed to varying degrees. This article introduces ...
NVIDIA Open-Sources LongLive 2.0: NVFP4 Quantization Enables 5B Models to Generate Long Videos in Real-Time, VRAM Usage Cut in Half, Speed Doubled
NVIDIA's research team recently officially open-sourced LongLive 2.0, a system based on NVFP4 (4-bit floating-point) quantization for
Colibri: Netizen Successfully Runs 744-Billion-Parameter GLM-5.2 on a Laptop with 25GB RAM
Recently, an independent developer created an inference engine called Colibri (colibrì) written entirely in C, successfully running on just 25GB
Berkeley Launches PixelRAG: Enabling AI to “See” Web Pages, Surpassing Pure Text RAG Accuracy by 18%
Traditional web crawling technology is about to undergo a major revolution. Traditional crawling techniques have issues such as being easily ...