#open-weight — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #open-weight, aggregated by home.social.
-
@alessiopomaro arriva la paura verso gli open-weight.
-
#KI #MistralAI
(1/n)"Überraschendste"?!?
Spätestens seit der Veröffentlichung des #TheEconomist-Interviews mit #ArthurMensch am 15.07.26 war dies der Öffentlichkeit bekannt. Dem Economist offenbar schon länger.https://www.economist.com/podcasts/2026/07/15/how-europe-can-compete-on-ai
"Der 👉überraschendste 👈Schritt: #Mistral bietet künftig auch #OpenWeight-Modelle anderer Entwickler über seine Plattform an. Den Anfang macht GLM-5.2 des #chinesischen KI-Labors Z.ai. Damit verdient das Unternehmen an fremden Modellen,...
-
🚨 Breaking news! Qwen 3.8-27B is set to go "openweight" in 2 days, whatever that means. 🙄 Meanwhile, Hugging Face has mastered the fine art of losing important pages in their labyrinthine digital jungle. 🌴🔍 The real question is, will they find them before Qwen makes its grand entrance? Place your bets, folks! 💸
https://huggingface.co/Qwen/Qwen3.8-27B #Qwen3.8 #HuggingFace #Openweight #TechNews #DigitalJungle #HackerNews #ngated -
Qwen 3.8-27B goes openweight in 2 days
https://huggingface.co/Qwen/Qwen3.8-27B
-
It's absolutely bonkers how far the #ai landscape moves. I try to keep up with the #openweight models... But it's every week a new model or new paper comes out. 🫠
Using is one thing, but building these things is a whole different level of insanity. 😅
-
Interesting to see Mark Zuckerberg's Meta pushing the open-weight AI model "Muse Glimmer" similar to the Chinese companies, but in contrast to most US ones.
llama.cpp integrations available at launch [1]
I also like this quote: "The notion AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic" [2]
Ref:
- [1] https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model
- [2] - https://www.myjoyonline.com/meta-launches-new-ai-model-as-zuckerberg-champions-open-weight-push
- Essay from Mark: https://www.meta.com/thefutureisforeveryone/
#AI #meta #openWeight #museGlimmer -
#AMD #Halo2 is pulling 200 tokens per second in your home.
#Opus5 about 55, 150 tokens per second downhill with a wind in its back.
For $10K, you can get a home #AI brain that can run 120Billion #OpenWeight #llm model that does not need a #Datacentre
https://www.amd.com/en/products/processors/desktops/ryzen/ryzen-ai-halo.html
-
CW: Local, open-source, (mostly) open-weight, sandboxed AI workspace 3 - Changes/musings
I'm pretty certain I'll remove the stability-matrix from the setup as I primarily used it for research/learning (Note: I wanted to see how it works and what parts were required, how they fit together....)
Now it's a resource hog for a piece of tech I don't need. There is enough open-source resources available and frankly am a little uncomfortable running anyway?
I also need to refine the specific model param sizes and quant configuration across my cluster, because right now some tasks are being done by models 10x the size required for that level of complexity.
Just downscaling the embedder to nomic from qwens already efficient one was a 90% VRAM usage cut for that task 🙃
-
CW: Local, open-source, (mostly) open-weight, sandboxed AI workspace 2
Usage is categorised into:
- Chat (quick, simple queries)
- Cowork (specific resource access, approval required)
- Agentic (sandboxed, no approvals required)
With all work:
- Organised into projects with dedicated, sandboxed workspaces
- Branching plan structure (see below)
- Chat is it's own project
- Conversational Memory and user profile is limited to per project
Plans get structured as tree's with an index at the root to allow the AI to crawl through it, i.e.
Plan A
| |
Plan B Plan C
| | |
Plan D Plan E Plan F -
CW: Local, open-source, (mostly) open-weight, sandboxed AI workspace
Summary:
- Server: LMStudio
- Client: Hermes + ByteBuddies (closed alpha)
- Sandboxing: double-layered of own network/router with dedicated machine + docker containerization.
- Architecture: MoM (5 models, 9 roles)
- Models: Mixture of Mistral (OCR, TTS) and Qwen (rerank, classify, code, main, agent), Nomic (embedding, Gemma (translate, summarize)
- Tool-count: 14 packs (±130 total)
- Skills-count: ±80 built-in+community and ±40 customs
- Assistant prompts: 18 (behaviour, role)
- CLI: OpenCode
- Image-Gen & Edit: StableDiffusion (Flux 2 small) via stabilitymatrix
Gateway and agentic assignments are handled via Hermes similarly to OpenClaw, but not shit.
Also been looking into Bonsai because their new 27b model, especially with tertiary enabled (for macs/m-chips) looks really promising.
-
My completely local/on-prem/open-weight AI setup is now able to describe, classify and tag images.
Now to clone a subset of my decade or so of images to see how well it does on a larger sample-size and with real-life auto-sorting :3
-
"On 17 July, China delivered its answer to Pax Silica by unveiling the World Artificial Intelligence Cooperation Organisation (WAICO) in Shanghai. Bringing together 29 nations across Eurasia, Africa, and Latin America - including its surviving socialist bastions of Cuba, Venezuela, and Nicaragua - WAICO unites the very global majority that's rendered surplus by the West's techno-utopian fantasies. Its first flagship initiative, announced by Xi Jinping, is to freely disseminate the Al-powered meteorological warning system MAZU. A clearer contrast with the paranoid securitisation of Pax Silica could hardly be imagined."
via @/sovereign_media_ on IG
https://www.instagram.com/reel/DbDoPX2o48m/#waico #ai #paxsilica #china #internationalism #opensource #openweight
-
#openweight models are to #Ai...
... What printing press was to #literacy -
It is a trivial skill to install an #abliterated model.
Literally takes 2 steps.
You don't need #huggingface spacesOn the other hand the US techbros want to #regulateai in the worst possible way.
Ban #openweight models and make their in-house, commercial models the only legal #AiTake heed that you do not carry the water for the #broligarchy with your #csam outrage
I used to be an internet #freespeech activist (long before the term became a dog whistle for #fascists), much of the unfree Internet and oppressive #internet legislation of today was led by the "Think of the children!" narrative.
I don't know, maybe a good idea not to do the stupid thing AGAIN.
-
At $dayjob, I have to use #claude and #gemini. And I do.
But at home, I use solely #openweight models. However, these models have gotten advanced. Enough so I prefer them. 😎 Like they seem to follow directions better and stay more focused than the frontier ones.
-
Бенчмарки врут? Бизнес мигрирует на открытые веса, а FDA и ФЗ заставляют уходить на свой инференс: ML-дайджест
Если водить спорткар, то только на идеальном глянцевом треке. Тогда он всегда будет показывать рекордное время. Я думаю, примерно так топ-менеджмент ИИ-компаний и описывает работу своих моделей инвесторам. Проблема в том, что бизнес-процессы — это дорога с ямами. Долгое время индустрия закрывала глаза на то, как именно LLM бьют рекорды на лидербордах. Но эпоха слепого доверия корпоративных данных чужим «черным ящикам» по ту сторону API подходит к концу. Или нет? В новом дайджесте обсуждаем масштабный переезд enterprise-сектора на открытые веса, суверенное железо и жесткую валидацию моделей на грязном реальном трафике.
https://habr.com/ru/companies/selectel/articles/1064784/
#selectel #ai #ии #llm #llmагент #itкомпании #OpenAI #sota #openweight #иизаконы
-
https://winbuzzer.com/2026/07/29/anthropic-rejects-open-weight-ban-proposes-ai-tests-xcxwbn/
Anthropic rejects an open-weight AI ban by the US & proposes capability-based safety tests, chip controls for China & action against large-scale model distillation.
#AI #Anthropic #DarioAmodei #OpenWeight #AIRegulation #AISafety #AIChips #ChinaAI #USExportControls
-
Numerama: C’est parti : les #poids de #Kimi #K3, plus gros #modèle #IA #ouvert au #monde, sont en ligne
#openweight is NOT #opensource
#opensource mattersEventhough #openweight is better than #closedweight..
-
"Reports suggest some US officials are considering banning the use of Chinese open-weights models by US companies. [...] #Openweight models [...] potentially present a higher risk than closed models, as it's very difficult to apply guardrails to them/monitor their usage."
Meanwhile reports PROVE #OpenAI can't control their models, #attacking companies with impunity.
It's very difficult to apply #guardrails on massive trillion dollar companies.
Scum #AI conglomerates.
-
Šéf Anthropicu Dario Amodei zveřejnil 27. července stanovisko k open-weight AI modelům. Reaguje na debatu posledních dnů, kdy americké úřady údajně zvažují zákaz používání čínských open-weights modelů americkými firmami a řada technologických firem se proti tomu ohradila otevřeným dopisem. Objevila se i obvinění, že o zákaz usiluje sám Anthropic kvůli ochraně […]
https://zdrojak.cz/zpravicky/anthropic-zakaz-open-weight-modelu-nechceme/ -
Anthropic CEO in a letter urging ban on model distillation. Same old story
- Still thinks "Distillation" is something different to scraping the Internet.
- Still thinks Chinese "authoritarian regime" is different to their own "authoritarian regime".
- Still technically powerless from being scraped, just like any other server on the internet, so they need laws to protect their specific server. -
The second MoonShot AI shoe drops....
Moonshot AI will make the weights of its Kimi K3 model available for unrestricted public download today. K3 contains 2.8 trillion parameters.
Founder Yang Zhilin has said the company wants to grow its user base through openness and broader availability than competing proprietary systems. https://qz.com/moonshot-ai-kimi-k3-open-weights-download-072726 #YangZhilin #MoonShotAI #AI #OpenWeight #OpenSource #OpenSourceAI #Kimi #K3 #KimiK3 #LLMs #FrontierModels
-
RT @chadwahl: Versandcontainer voller B300s + offene Weight-Modelle + Palantir AIP, die überall auf der Welt laufen, verwaltet von Apollo 🔥 Wir haben dies letzten Herbst als Teil unserer Referenzarchitektur für ein souveränes AI-OS vorgestellt. Alle reden darüber, wir liefern bereits. Chad Wahlquist (@chadwahl) Wir werden so viele offene Weight-Modelle vor Ort einsetzen — https://nitter.net/chadwahl/status/2080846140396720265#m
mehr auf Arint.info
-
"American labs need to release frontier-grade open-weight models under licenses that startups can actually build on. There has been progress. NVIDIA’s Nemotron models are commercially usable under NVIDIA’s own permissive license. Thinking Machines released Inkling under Apache 2.0, as did OpenAI with gpt-oss and Google with Gemma 4. But OpenAI’s and Google’s strongest models remain closed, as do those from most American frontier labs.
Use procurement to create an open market
The government should use procurement to create demand for portable, interoperable systems rather than permanent dependence on one API vendor. The Department of Defense has done this before. Platform One provides open source tools and enterprise products that different military programs can build on. The same playbook can accelerate innovation around open-weight models.Build the rest of the stack
American companies need to build the rest of the stack. Startups can customize and extend the models, embed them into products, and provide the serving, tooling, support and operational layers. Our leading silicon companies will keep improving the hardware. Hyperscalers and neoclouds can serve the models and their ecosystems.Set standards instead of banning models
Safety is the strongest argument for restrictions, but a blanket ban is too blunt and would sacrifice access to the entire ecosystem. A better approach is independent testing and standards for frontier models. The analogy isn’t exact: Kubernetes conformance tests compatibility, not safety. But the governance model is useful. Demis Hassabis has proposed a US-led independent standards body along those lines.America should not respond to open Chinese models by building a wall around its own developers. We should run the models ourselves, tear them apart, benchmark them, improve on them and build better American alternatives."
https://tobi.knaup.me/2026-07-25-open-weight-ai-is-having-its-kubernetes-moment/
#AI #GenerativeAI #Kubernetes #USA #China #OpenSource #OpenWeight #LLMs
-
🦾 China’s Kimi K3 and the rise of open-weight AI models
｢ OpenAI’s head of strategic futures, argued that a world dominated by open-weight models could lead to “full AI communism”—a future he described as “a dystopian hellscape.” ｣
https://www.scientificamerican.com/article/china-kimi-k3-and-the-rise-of-open-weight-ai-models/
-
I do find the #openweight discussions quite interesting. As a #floss developer I generally favour openness although I'm never quite clear what the training inputs are. They could be the product of #distillation attacks but that's basically hoovering up the internet with extra steps which doesn't seem so different from the big #frontier models anyway: https://www.youtube.com/watch?v=YP73B9D20V4 #youtube #fireship
-
The Corrupt, Xenophobic Hysteria Behind The ‘TikTok Ban’ Will Soon Be Mirrored Across U.S. AI Policy