金のニワトリ (@gosrum)

GLM-5.1의 ts-bench 벤치마크 결과를 공유했다. 다른 로컬 LLM도 만점을 받은 적은 있지만, GLM-5.1은 N=3 조건에서 일관되게 만점을 기록한 첫 로컬 LLM이라고 강조한다.

https://x.com/gosrum/status/2041709112661008859

#glm #benchmark #localllm #tsbench #llm

早速試したのでブログにまとめた

はてなブログに投稿しました
GitHub Copilot CLIでローカルLLMを使って完全オフラインでコードを書かせる方法 - await wakeUp(); https://sublimer.hatenablog.com/entry/2026/04/08/184248

#はてなブログ #GitHub_Copilot_CLI #LocalLLM

MekaHime (@MekaHimeAI)

AI waifu 'Amika' 개발 비용이 현재까지 약 2.5만 달러라고 소개했다. 자체 STT·TTS와 커스텀 동적 프롬프팅 시스템을 사용하며, 로컬 LLM만으로 800ms 미만 응답 속도를 구현한 사례로, 실시간 대화형 AI 제품/애플리케이션 관점에서 흥미롭다.

https://x.com/MekaHimeAI/status/2041213151526703370

#aiwaifu #stt #tts #localllm #prompting

Local AI! Mini-LLM!

Currently, a large portion of the work can be done on an ancient laptop running Linux Mint, 16GB RAM, a 4B-Model and LLMStudio.

Who needs gigantic data-centers? Not I! ;0)

It's not the size of your tech that matters ... it's what you do with what you got

#AnotherFineMyth #AFMFanFilm #OpenSource #LocalLLM #AIArt

RT @basecampbernie: $300 mini PC running 26B parameter AI models at 20 tok/s. Minisforum UM790 Pro ($351) + AMD Radeon 780M iGPU + 48GB DDR5-5600 + 1TB NVMe. The secret: the 780M has no dedicated VRAM. It shares your DDR5 via unified memory. The BIOS says "4GB VRAM" but Vulkan sees the full pool. I'm allocating 21+ GB for model weights on a GPU with "4GB VRAM." The iGPU reads weights directly from system RAM at DDR5 bandwidth (~75 GB/s). MoE only activates 4B params per token = 2-4 GB of reads. That's why 20 tok/s works. What it runs: - Gemma 4 26B MoE: 19.5 tok/s, 110 tok/s prefill, 196K context - Gemma 4 E4B: 21.7 tok/s faster than some RTX setups - Qwen3.5-35B-A3B: 20.8 tok/s - Nemotron Cascade 2: 24.8 tok/s Dense 31B? 4 tok/s, reads all 18GB per token, bandwidth wall. MoE same quality? 20 tok/s. Full agentic workflows via @NousResearch Hermes agent with terminal, file ops, web, 40+ tools, all against local models. No API keys. Just a box on your desk. The RAM is the pain right now. DDR5 prices 3-4x what they were a year ago. But the compute is free forever after you buy it. @Hi_MINISFORUM @ggerganov llama.cpp + Vulkan + @UnslothAI GGUFs + @AMDRadeon RDNA 3. Fits in your hand. #LocalLLM #Gemma4 #llama_cpp #AMD #Radeon780M #MoE #LocalAI #AI #OpenSource #GGUF #HermesAgent #NousResearch #DDR5 #MiniPC #EdgeAI #UnifiedMemory #Vulkan #iGPU #RunItLocal #AIonDevice

Mehr auf Arint.info

#agent #API #GGUF #llama #LocalAI #OpenSource #Qwen3535 #arint_info

https://x.com/basecampbernie/status/2040326984446935059#m

Code's Local Limit: When Big Models Break Small Machines

Running large language models for coding locally is limited by RAM. Users need more memory for bigger models, affecting small computer use.

#LocalLLM, #CodingAI, #RAMLimit, #ComputerHardware, #AIonPC

https://newsletter.tf/local-llm-coding-ram-limit-small-computers/

Using large language models for coding on your own computer needs a lot of RAM. If your computer has less than 16GB of RAM, you might not be able to run bigger models for coding.

#LocalLLM, #CodingAI, #RAMLimit, #ComputerHardware, #AIonPC
https://newsletter.tf/local-llm-coding-ram-limit-small-computers/

Just tried out Gemma 4 - E3B locally on my pixel phone. Using Google Edge Gallery with network permissions disabled (GrapheneOS)

It understands audio. Maybe image works too. Speed is decent. As long as prompts are simple and clear, i think its useful.

Not sure about battery consumption. But i bet for 80% of cases we dont need data center. It might not program but can tell u how to color SVG when u are offline.

#localllm #gemma

Ollama가 Apple의 ML 프레임워크 MLX 기반으로 Apple Silicon(M5/M5 Pro/M5 Max)에서 미리보기로 가속됩니다. Qwen3.5-35B-A3B에서 prefill·decode 속도 크게 향상되고 NVFP4 양자화로 생산 환경과 동등한 품질 유지가 가능해졌습니다. 캐시 재사용·스마트 체크포인트·스마트 삭제로 응답성·메모리 효율 개선. Ollama 0.19 공개(통합메모리 32GB 권장).

https://ollama.com/blog/mlx

#applesilicon #mlx #nvfp4 #localllm #performance