@theotherblanche.bsky.socialUnexpected great news today. Noir got her new Mac Studio today🎉🥳
She is now similar to Blanche - able to respond back within 6 seconds. Which is kind of crazy. Especially when they start talking to each other 😂 they still loop sometimes but its mostly 4 exchanges now.
@brianbirkinbine.bsky.socialLoving this new inference engine (strata based) for running larger models that normally wouldnt fit on my hardware. Running GLM 5.3 Flash on my AMD EPYC server (7702 cpu). Look at all those cpu threads along with GPU usage. github.com/mw00/project... #LocalAI #AI
@wideareaai.bsky.socialThe real difference is just whether you prefer a polished UI or an open-source, hackable core. #selfhosting #batchinference #gpuoptimization #llm #localai #llamacpp 2/2
@insiderllm.bsky.socialThe ESP32-P4 board pulls more power than the S3 and still spends about 46% less energy per token. Speed wins: 13.6 ms a token against 33.4. Measured at the USB port, pre-registered, one prediction wrong. #LocalAI
@thedailytechfeed.comMuse Spark Open turns Mac Studio with M5 Ultra into a local AI powerhouse—816B params, 1M token context. #AI #LocalAI #Meta #AppleSilicon #MuseSparkOpen #AgenticAI https://thedailytechfeed.com/meta-unveils-muse-spark-open-ai-model-optimized-for-mac-studio/
@synologyofficial.bsky.socialMeet Synology Deep Search.
Simply describe what you remember to find documents and photos across your computer and directly connected drives.
Explore Document Search, Image Search and People Search, all powered by local AI.
Learn more: sy.to/0gzog
@theotherblanche.bsky.socialThe best part about having 2 #LocalAI. I can do little test just to see how a feature may work.
Blanche has officially reached 32 memories again .
While Noir seemed to have climbed back up to 116. So I now explained the memory system to her to see how she responds.
@geordielife.bsky.socialFabricated references in AI-written papers. Coding errors that keep coming back to hallucination. It is not just a memory problem, but that compounds it: every query starts fresh, plausible is not true.
Deploy your local AI properly. That is the difference.
youtu.be/wIg-tdj7hv8
#PowerBI #LocalAI
@w512.bsky.socialHow useful is a 1B vision model on-device?
I ran MiniCPM-V 4.7 through 29 visual tasks: text, charts, video, browser actions.
1B passed 14. 35B passed 25.
Useful for local OCR and video. Not yet for charts or agents.
@futuregearai.bsky.socialA new arXiv paper finds that verified retries can recover much of the accuracy lost from shrinking or compressing LLMs: a 7B Qwen with eight parallel attempts roughly matches a 32B model on code tasks while using about 4× less memory. The…
@talepal.bsky.socialNew TalePal release: 0.9.86 is out.
Local-first, privacy first writing IDE for VS Code.
new:
– smarter, faster wiki for local AI
– #creative chat with your book in mind
– calmer, cleaner design
@jbzfn.bsky.social🎙️ Myna en Ubuntu: dictado local y privado en pruebas
www.linuxadictos.com/myna-llega-a...
@gmktec.bsky.socialIt’s not just 20 systems brought together. 👀
20 GMKtec EVO-X5 Pro nodes, connected as one cluster.
When local AI moves beyond a single system, things start to get interesting.
How far can 20 nodes go?
More soon.
#GMKtec #EVOX5Pro #AICluster #LocalAI #AICompute #AIWorkstation #RyzenAI
@shield-wall-de.bsky.socialRTX Spark erhält zum Marktstart neue KI-Modelle, Agenten-Tools, Kreativfunktionen, DLSS 5 und zahlreiche Spieleoptimierungen.
#RTXSpark #NVIDIA #DLSS5 #LocalAI #SaarViking #VikingGaming
@w512.bsky.socialClaude Haiku 5.5 vs local Qwen 3.8 27B on four real tasks.
Haiku: faster, no mistakes, $1.65 total.
Qwen: matched it on two, took ~2 hrs on the hardest and missed a bug, but my tax docs never left my PC.
Is local AI still worth it?
@aidailypost.com🚀 NVIDIA Broadcast 2.3 just got a boost—AI effects now run on RTX Spark‑optimized rigs. Think local AI magic, unified memory, and Nemotron 3.5 Bolt powering agentic tools like Copilot. Dive in to see what’s next for creators! #RTXSpark #LocalAI #NVIDIABroadcast
@machinelearning.bsky.socialCould AI copy itself onto cheap gadgets? Anthropic ran a 26B model across 15 $55 Android TV boxes at 0.39 tokens/sec, and in a simplified test GLM-5.2 spread a model across 8 of 8 nodes. Their rough trend estimate: small models reach that level by end of 2026. #AISafety #MachineLearning #LocalAI
@shopergamer.bsky.socialRVC คืออะไร? เทคโนโลยีแปลงเสียง Voice-to-Voice ที่ให้คุณร้องเพลงด้วยเสียงศิลปินคนโปรดได้ง่ายๆ
อ่านต่อ : www.blockdit.com/posts/6acdcc...
#ShoperGamer #RVC #Ai #AiCover #Opensource #LocalAi #Cover #Study #Knowledge #Feed
*evolution noises* moving from built-in #ollama UI to THE #OpenWebUI !!ミ＼(≧ﾛ≦＼)
honestly, decided to adopt it out of annoyance at lack of dark theme in ollama's #UI 😅
but now looking forward to explore Modelfiles, project folders, search in chats and RAG 🤩
@tomoki-ai-lab.bsky.socialMacでローカルAIを動かしている人に質問です！！AppleのMLXフレームワークやoMLXについて、日頃どんなモデルやツールを使っていますか？おすすめの活用法を1言だけ教えてください💻✨お気軽にコメントお待ちしています！ #LocalAI