#qwen3 — Public Fediverse posts

🚀 Wow, "Qwen 3.8" is now hot on the heels of "GPT-5.5 Pro" in the prestigious field of... *reasoning prefills* 🤯. Because, let's face it, who doesn't love a good prefilled reasoning session from the vast wilderness of #GitHub gists? 😜
https://gist.github.com/wsxiaoys/e0286dc6bb624ff5fdf49e7f4c528ba3 #Qwen3.8 #GPT5.5 #reasoningprefills #gists #AItechnology #HackerNews #ngated

🚀 Wow, "Qwen 3.8" is now hot on the heels of "GPT-5.5 Pro" in the prestigious field of... *reasoning prefills* 🤯. Because, let's face it, who doesn't love a good prefilled reasoning session from the vast wilderness of #GitHub gists? 😜
https://gist.github.com/wsxiaoys/e0286dc6bb624ff5fdf49e7f4c528ba3 #Qwen3.8 #GPT5.5 #reasoningprefills #gists #AItechnology #HackerNews #ngated

🚀 Wow, "Qwen 3.8" is now hot on the heels of "GPT-5.5 Pro" in the prestigious field of... *reasoning prefills* 🤯. Because, let's face it, who doesn't love a good prefilled reasoning session from the vast wilderness of #GitHub gists? 😜
https://gist.github.com/wsxiaoys/e0286dc6bb624ff5fdf49e7f4c528ba3 #Qwen3.8 #GPT5.5 #reasoningprefills #gists #AItechnology #HackerNews #ngated

🚀 Wow, "Qwen 3.8" is now hot on the heels of "GPT-5.5 Pro" in the prestigious field of... *reasoning prefills* 🤯. Because, let's face it, who doesn't love a good prefilled reasoning session from the vast wilderness of #GitHub gists? 😜
https://gist.github.com/wsxiaoys/e0286dc6bb624ff5fdf49e7f4c528ba3 #Qwen3.8 #GPT5.5 #reasoningprefills #gists #AItechnology #HackerNews #ngated

🚀 Wow, "Qwen 3.8" is now hot on the heels of "GPT-5.5 Pro" in the prestigious field of... *reasoning prefills* 🤯. Because, let's face it, who doesn't love a good prefilled reasoning session from the vast wilderness of #GitHub gists? 😜
https://gist.github.com/wsxiaoys/e0286dc6bb624ff5fdf49e7f4c528ba3 #Qwen3.8 #GPT5.5 #reasoningprefills #gists #AItechnology #HackerNews #ngated

Qwen 3.8 follows GPT-5.5 Pro reasoning prefills

https://gist.github.com/wsxiaoys/e0286dc6bb624ff5fdf49e7f4c528ba3

Comments: https://news.ycombinator.com/item?id=49630026

#HackerNews #Qwen3.8 #GPT5.5 #Pro #Reasoning #AIUpdates #TechNews

Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses

https://quesma.com/blog/qwen38-27b-quantizations-benchmarked/

Comments: https://news.ycombinator.com/item?id=49611128

#HackerNews #Benchmarking #Qwen3.8 #27B #quantizations #4bit #1bit #AIperformance #Qwen38

Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses

https://quesma.com/blog/qwen38-27b-quantizations-benchmarked/

Comments: https://news.ycombinator.com/item?id=49611128

#HackerNews #Benchmarking #Qwen3.8 #27B #quantizations #4bit #1bit #AIperformance #Qwen38

Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses

https://quesma.com/blog/qwen38-27b-quantizations-benchmarked/

Comments: https://news.ycombinator.com/item?id=49611128

#HackerNews #Benchmarking #Qwen3.8 #27B #quantizations #4bit #1bit #AIperformance #Qwen38

Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses

https://quesma.com/blog/qwen38-27b-quantizations-benchmarked/

Comments: https://news.ycombinator.com/item?id=49611128

#HackerNews #Benchmarking #Qwen3.8 #27B #quantizations #4bit #1bit #AIperformance #Qwen38

Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses

https://quesma.com/blog/qwen38-27b-quantizations-benchmarked/

Comments: https://news.ycombinator.com/item?id=49611128

#HackerNews #Benchmarking #Qwen3.8 #27B #quantizations #4bit #1bit #AIperformance #Qwen38

RT @OrcaRouter: Jemand hat unabhängig 8 verschiedene Qwen3.8-27B-Modelle ohne Zensur oder Abliteration benchmarkt.

mehr auf Arint.info

#Abliteration #AIResearch #Benchmark #LLM #OrcaRouter #Qwen3 #arint_info

https://x.com/OrcaRouter/status/2096828627807768836

RT @imnotchalk: Qwen3.8 27B ist jetzt im @cerebras Shared Tier verfügbar mit einer Geschwindigkeit von ~1500 Token pro Sekunde. 🚀

mehr auf Arint.info

#AI #Cerebras #LLM #MachineLearning #Qwen3 #TechNews #arint_info

https://x.com/imnotchalk/status/2095637567979114654

🚀🎉 Big news, everyone! The latest and greatest Qwen 3.8 27B is out on #Cerebras, promising to process text faster than you can say "overhyped tech jargon." Just navigate through a labyrinth of buzzwords and obscure documentation to find out how to pay for it - because who doesn't love a treasure hunt through a digital maze? 💸🔍
https://inference-docs.cerebras.ai/models/overview #Qwen3.8 #TechBuzz #DigitalMaze #TreasureHunt #HackerNews #ngated

Qwen 3.8 27B available on Cerebras at 1500 tok/SEC

https://inference-docs.cerebras.ai/models/overview

Comments: https://news.ycombinator.com/item?id=49554520

#HackerNews #Qwen3.8 #Cerebras #AI #Model #1500tok #SEC #MachineLearning

Another #localLLM #Qwen3.8 on 2x12GB VRAM test:

Asked to write a browser-based #FTIR DX-data-to-graph render widget. It did so in less than 2 minutes. 🤯

Attached are 3 graphs: 2 are from official lab tools - one generated by #GenAi code.

Can you guess which one is which?

Running 104GB Qwen3.8-Flash-Next on 48GB Mac with at ~12 tok/s

https://github.com/carloslfu/slotstream

Comments: https://news.ycombinator.com/item?id=49524447

#HackerNews #Running #Qwen3.8-Flash-Next #Mac #Performance #AI #104GB

Lokale KI erstmals auf Cloud-Niveau?

https://www.youtube.com/watch?v=HOhw1ywkMUU

#ki #lokaleki #qwen #qwen3

RT @danieltvela: Ich habe Qwen3.8-Flash-Next auf einem PRO 6000 mit ~56 GB DRAM am Laufen.

mehr auf Arint.info

#AI #MTP #Performance #Prefilling #Qwen3 #vllm #arint_info

https://x.com/danieltvela/status/2093999978473542089

Interesting: i gave claude and Qwen 3.5 9B the same problem to solve do a small home assistant automation. Qwen was running on my 5.5 year old Laptop (8 core Ryzen, 32gig Ram no useable GPU(at least i think so)) claude was done in 20sec, my Laptop needed 240sec but the endresult was nearly identical and correct. Also the annotations were nearly identical. Okay it is slow but better use my green power here than using claude which site in some coal powered AWS etc datacenter
#ai #janai #qwen3.5

Just ran Qwen3.8-Flash-Next locally with Unsloth UD-IQ3_XXS quantization. Set thinking mode to XHIGH.

@simon's classic prompt: Generate an SVG of a pelican riding a bicycle.

Local AI is getting scarily good.

#LocalAI #Qwen3 #Unsloth #GenerativeAI

🚀 Oh look, another tech bro's magnum opus: a riveting tale of running #Qwen3.8 on a Mac Studio, complete with thrilling RAM usage stats and a 20-minute "gotcha" 🙄. Truly, it's the War and Peace of terminal commands! 🍿
https://terminalbytes.com/run-qwen-3-8-27b-locally/ #techbro #tales #MacStudio #RAMusage #terminalcommands #WarandPeace #HackerNews #ngated

RT @OrcaRouter: Qwen3.8 Flash Next. Unzensiert. Jetzt in NVFP4. 🐳

mehr auf Arint.info

#AI #HuggingFace #MachineLearning #NVFP4 #OpenSource #Qwen3 #arint_info

https://x.com/OrcaRouter/status/2093233719473746065

RT @0xBakeer: Qwen3.8-Flash-Next auf einer einzelnen DGX Spark: bis zu 97 Tokens pro Sekunde.

mehr auf Arint.info

#AI #DGXSpark #MachineLearning #NLP #OpenSource #Qwen3 #arint_info

https://x.com/0xBakeer/status/2092709567503229015

🚀⚡ "Qwen3.8-Flash-Next" might be the next big thing, if only anyone understood what 'Artificial AnalysisArtificial AnalysisKArtificial' is supposed to mean. 🤔 With intelligence and speed measured in units of gibberish, one can only wonder if the biggest #innovation is in how many #buzzwords they can fit on a page. 💸
https://artificialanalysis.ai/models/qwen3-8-flash-next #Qwen3.8FlashNext #ArtificialIntelligence #TechTrends #HackerNews #ngated

Qwen3.8-Flash-Next Intelligence, Performance and Price Analysis

https://artificialanalysis.ai/models/qwen3-8-flash-next

Comments: https://news.ycombinator.com/item?id=49461862

#HackerNews #Qwen3.8 #Flash #Next #Intelligence #Performance #Price #Analysis #AI #Technology #Innovation

RT @pupposandro: Mit Spannung verkünde ich, dass die Lucebox-Engine nun Qwen3.8-27B auf einer einzelnen AMD Radeon AI PRO R9700 (32 GB, RDNA4) mit dem DFlash2-Block-Diffusions-Drafter von z-lab ausführt:

mehr auf Arint.info

#AI #AMD #Lucebox #MachineLearning #Qwen3 #RDNA4 #arint_info

https://x.com/pupposandro/status/2092614392202313848

🔍🎨 So, it seems Qwen 3.8 is the new Picasso of #AI, with its 27 billion parameters ready to dazzle—or confuse—you with a dizzying array of credits and models that sound like a fruit salad 🍌🍎. Who knew API testing involved so many Nano Bananas and Seedreams? At least now we know #ByteDance isn't just for TikTok dances; it’s also for pricey AI models. 💸💃
https://imageat.com/models/qwen-3-8-27b-uncensored #Art #Qwen3.8 #APItesting #TechTrends #HackerNews #ngated

Qwen 3.8 27B Uncensored – Testing the Uncensored Qwen Model

https://imageat.com/models/qwen-3-8-27b-uncensored

Comments: https://news.ycombinator.com/item?id=49456134

#HackerNews #Qwen3.8 #Uncensored #AI #Model #Testing #MachineLearning #TechNews

🚀 Oh wow, yet another #architecture promising "ultimate cost-efficiency"—because we all know that's never been done before! 🙄 Qwen3.8-Flash-Next sounds like the latest iOS update nobody asked for, but hey, at least it's got a snappy name! 😂
https://qwen.ai/blog?id=qwen3.8-flash-next #ultimatecostefficiency #Qwen3.8FlashNext #techhumor #innovation #HackerNews #ngated

Qwen3.8-Flash-Next: A New Architecture, Towards Ultimate Cost-Efficiency

https://qwen.ai/blog?id=qwen3.8-flash-next

Comments: https://news.ycombinator.com/item?id=49448210

#HackerNews #Qwen3.8 #Flash #Next #Ultimate #Cost #Efficiency #New #Architecture

RT @UnslothAI: Jetzt kannst du Qwen3.8-27B kostenlos mit unserem Notebook feinabstimmen! 🔥

mehr auf Arint.info

#AI #DeepLearning #FineTuning #MachineLearning #OpenSource #Qwen3 #arint_info

https://x.com/UnslothAI/status/2092255772214509698

@OliDietzel: Noch 8 Stunden bis zum Release von #Qwen3.8-Flash-Next x.com/ModelScope2022…

mehr auf Arint.info

#AI #FlashNext #ModelScope #Qwen3 #Release #arint_info

https://x.com/OliDietzel/status/2092503972552577279

📰 Oh, look! Another AI model with a name that sounds like a bad sci-fi sequel: "Qwen 3.8-Flash-Next" is dropping tomorrow! 🎉 Because what we really needed was an even more bloated version of "ModelScope"—because, as we know, size totally equals intelligence, right? 🙄 #InnovativeNaming #MoreIsMore
https://modelscope.cn/models/Qwen/Qwen3.8-Flash-Next #AIModels #Qwen3.8 #FlashNext #ModelScope #TechHumor #Innovation #HackerNews #ngated

Qwen 3.8-Flash-Next releasing tomorrow (125B a6B)

https://modelscope.cn/models/Qwen/Qwen3.8-Flash-Next

Comments: https://news.ycombinator.com/item?id=49432317

#HackerNews #Qwen3.8FlashNext #AI #Models #Release #Tomorrow #Tech #News #ModelScope

RT @ciruai: Das beste Qwen3.8 27B-Modell für AMD Strix Halo ist da. In Zusammenarbeit mit Pete Hopton und dem Kairic.ai-Team präsentiere ich: Qwen3.8-27B-IU4-KAIRIC-EDGE. Unserer Kenntnis nach ist dies die weltweit erste öffentliche Veröffentlichung eines 27B-LLM mit einer beschleunigten nativen IU4-Inferenzspur, speziell für AMD Strix Halo (gfx1151) entwickelt. Warum ist IU4 wichtig? Die meisten 4-Bit-Modelle speichern zwar Speicher, konvertieren Gewichte jedoch weiterhin in breitere Formate während der Ausführung. Unsere IU4-Spur hält Vier-Bit-Ganzzahldaten im aktiven Berechnungspfad und mappt sie direkt auf RDNA 3.5-Integer-Hardware. Das reduziert den Speichertraffic und verwandelt Quantisierung von einem Speicherformat in eine echte Hardware-Ausführungsstrategie. Der abgestimmte IU4-Operator erreichte 104,66 TOPS – 1,94× FP16 und 1,93× IU8. Dies eröffnet einen Weg zu schnellerer, effizienterer lokaler KI, ohne dabei die Modellqualität zu opfern. Ich kann es kaum erwarten zu sehen, was @Italianclownz und andere Zauberer damit anstellen. Angetrieben von Prompt Forge + Dual View, Kairic Edge-Beschleunigung, 262K Kontext: HumanEval-Metriken: • 47,73 tok/s Full-Suite TG • +85% gegenüber Unsloth Dynamic Q4 • +89% gegenüber Unsloth Dynamic Q6 • Schlug Unsloth Dynamic v3 q6 in Human Eval um 2 Fragen. Dies ist auch die erste Veröffentlichung, die von Kairic.ai angetrieben wird, einem neuen Unternehmen, das sich auf die gemeinsame Entwicklung von KI-Software rund um die Hardware konzentriert, auf der sie tatsächlich läuft. Modell, Benchmarks, Quelle, Build-Anleitung und Runner: huggingface.c…

mehr auf Arint.info

#AMDStrixHalo #KairicAI #LLM #LocalAI #Qwen3 #RDNA3 #arint_info

https://x.com/ciruai/status/2091004843720659229

RT @UnslothAI: Wir veröffentlichen neue Qwen3.8-27B GGUFs mit 10 % höherer Genauigkeit. Unsloth Dynamic V3 schlägt andere um 10 % auf Div-300, KLD und weiteren Benchmarks. Zudem veröffentlichen wir 1-Bit-Quantisierungen, die 77 % der Genauigkeit beibehalten. Läuft auf 8 GB RAM. Blog: unsloth.ai/docs/basics/dynam… GGUF: huggingface.co/unsloth/Qwen3…

mehr auf Arint.info

#AI #GGUF #MachineLearning #OpenSource #Qwen3 #Unsloth #arint_info

https://x.com/UnslothAI/status/2090103470015828184

Not a fan of Qwen 3.8 27B.

Overthinks too much => wasted tokens.

Has someone ran it without too much thinking? Does it perform well with that knob down, or letting it think all the way is required to be good?

#AI #ArtificialIntelligence #LLM #Qwen #Qwen38 #Qwen3 #Ollama #LlamaCCP #vLLM #LargeLanguageModels #LocalAI

Qwen 3.8 27B

https://huggingface.co/Qwen/Qwen3.8-27B-FP8

Comments: https://news.ycombinator.com/item?id=49299605

#HackerNews #Qwen3.8 #27B #AI #MachineLearning #HuggingFace #TechNews

🚀 Behold, the latest #AI model: Qwen3.8-2.4T, where numbers are random and acronyms are abundant, proving that the #tech industry’s true #innovation lies in finding new ways to baffle the uninitiated. 🤔 Hugging Face rolls out another #model, because why not throw more #jargon at you while you desperately try to figure out what “fp8” means. 🤷‍♂️
https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B-FP8 #Qwen3.8 #HuggingFace #HackerNews #ngated

🚨 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

🚨 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

🚨 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

🚨 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

🚨 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

Сжатие декодерных эмбеддеров: как ужать 8B до продакшена без потери recall

Декодерный эмбеддер 7–8B дает качество, но платит за него памятью, latency и деньгами. Разбираем все оси сжатия - int8, int4, binary + rescoring, PQ, MRL-усечение - на реальных замерах recall@10: где деградация мягкая, а где обрыв. С воспроизводимым кодом и Colab-ноутбуком под Qwen3

https://habr.com/ru/articles/1054930/

#сжатие_эмбеддингов #квантизация #эмбеддинги #embeddings #RAG #Qdrant #Qwen3 #binary_quantization #Matryoshka #retrieval

Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model

https://qwen.ai/blog?id=qwen3.6-27b

#HackerNews #Qwen3.6 #27B #Coding #AI #Model #Flagship #Technology

🎯 Supported models include #GPT-OSS-120B, #GPT-OSS-20B, #Llama4 Maverick, #Llama4 Scout, #Llama33-70B, #Llama31-8B, #KimiK2, #Qwen3-32B

🔧 Key features: deterministic inference for faster tool-using agents, cost-effective scaling, approved tool use with clear allowlists, seamless migration capability

📋 Ready-to-use cookbook tutorials with #BrowserBase #MCP, #BrowserUse #MCP, #Exa #MCP, #Firecrawl #MCP, #HuggingFace #MCP, #Parallel #MCP, #Stripe #MCP, #Tavily #MCP

How to vibe code for free: Running Qwen3 on your Mac, using MLX
https://localforge.dev/blog/running-qwen3-macbook-mlx
#ycombinator #Qwen3 #MLX #macOS #Apple_Silicon #Local_LLM #Localforge #Free_Code_Generation #Ollama #Local_AI #LLM_Agent