home.social

#gptoss — Public Fediverse posts

Live and recent posts from across the Fediverse tagged #gptoss, aggregated by home.social.

fetched live
  1. NVIDIA Blackwell Sweeps Every MLPerf 6.0 Benchmark With No Competition In Sight, While GB300 Systems Run Up to 60% Faster Than GB200 The latest MLPerf Training 6.0 benchmarks are in & NVIDIA ha...

    #Featured #News #Sticky #Blackwell #CoreWeave #DeepSeek #V3 #GB300 #GB300 #NVL72 #GPT-OSS

    Origin | Interest | Match
  2. Развиваем ИИ-платформу в 1С

    Статья о том, как мы пришли к ИИ для 1С и что из этого вышло. Материал получился про попытку собрать полноценную ИИ-платформу для работы с учетными данными. Расскажу, зачем нам понадобился MCP, как мы связали его с 1С:Шина, почему отдельно уперлись в права доступа и как в итоге устроили агента внутри 1С. Также покажу, где видим место для skills, RAG и OCR, и что пока не стали отдавать модели на самостоятельное выполнение.

    habr.com/ru/companies/cinimex/

    # #иипомощник #mcpserver #function_calling #агент #rag #ocr #права_доступа #локальные_модели #gptoss

  3. Миф о «равных весах»: что на самом деле скрывается внутри малых моделей

    Последние годы развитие LLM шло по пути экстенсивного масштабирования: считалось, что чем больше весов и данных, тем умнее модель. В индустрии даже сложилась жесткая классификация по количеству параметров: 7B, 8B, 32B. Она же создает иллюзию, что модели одной весовой категории обладают сопоставимыми аналитическими, генеративными и логическими характеристиками, что в корне противоречит современным эмпирическим наблюдениям. Но действительно ли «вес» модели все еще определяет ее качество в 2026 году? Или компактная архитектура способна конкурировать с гигантами, требующими H100 и сотни гигабайт VRAM? В этой статье мы не будем рассуждать об этом абстрактно: возьмем реальные модели из каталога FMC и посмотрим на практике, как размер влияет на качество reasoning, генерации и прикладную эффективность — и влияет ли вообще.

    habr.com/ru/companies/selectel

    #mlops #selectel #llmмодели #qwen #phi4 #mistral #gptoss #deepseek #ai #aiагенты

  4. Миф о «равных весах»: что на самом деле скрывается внутри малых моделей Последние годы развитие LLM шло по пу...

    #mlops #selectel #llm-модели #qwen #phi-4 #mistral #gpt-oss #deepseek #ai #ai-агенты

    Origin | Interest | Match
  5. Neste vídeo, eu testei diferentes modelos de inteligência artificial rodando localmente no Linux Mint utilizando o LM Studio.
    O objetivo foi verificar se essas IAs conseguem responder corretamente perguntas simples — desde fatos pouco conhecidos, como o primeiro vocalista do Iron Maiden, até informações mais populares, como quem criou o kernel Linux.

    Link: youtu.be/F8A2Yx4an_M

    #inteligenciaartificial #lmstudio #ialocal #ibmgranite #gptoss #googlegemma #qwen #rx550 #linux

  6. Neste vídeo, eu testei diferentes modelos de inteligência artificial rodando localmente no Linux Mint utilizando o LM Studio.
    O objetivo foi verificar se essas IAs conseguem responder corretamente perguntas simples — desde fatos pouco conhecidos, como o primeiro vocalista do Iron Maiden, até informações mais populares, como quem criou o kernel Linux.

    Link: youtu.be/F8A2Yx4an_M

    #inteligenciaartificial #lmstudio #ialocal #ibmgranite #gptoss #googlegemma #qwen #rx550 #linux

  7. New week, more slides: Run LLMs Locally

    Now with LFM 2 and new slides for using Transformers.js with WebGPU for Privacy Filter, Function Calling and Embeddings, running completely in your browser.

    codeberg.org/thbley/talks/raw/

    #ai #llm #llamacpp #stablediffusion #gptoss #qwen3 #glm #localai #gemma4 #nemotron #webgpu

  8. New week, more slides: Run LLMs Locally

    Now with LFM 2 and new slides for using Transformers.js with WebGPU for Privacy Filter, Function Calling and Embeddings, running completely in your browser.

    codeberg.org/thbley/talks/raw/

    #ai #llm #llamacpp #stablediffusion #gptoss #qwen3 #glm #localai #gemma4 #nemotron #webgpu

  9. TUTORIAL - Passo a passo: IA local no Linux com LM Studio

    Neste vídeo, você vai aprender como rodar inteligência artificial localmente no seu Linux usando o LM Studio.

    Se você quer mais controle, mais privacidade e rodar IA direto no seu computador, esse guia é pra você.

    Link: youtu.be/M7jR2BIuGyQ

    #debian #debian13 #lmstudio #ialocal #ibmgranite #gptoss #googlegemma #qwen #rx550 #linux

  10. TUTORIAL - Passo a passo: IA local no Linux com LM Studio

    Neste vídeo, você vai aprender como rodar inteligência artificial localmente no seu Linux usando o LM Studio.

    Se você quer mais controle, mais privacidade e rodar IA direto no seu computador, esse guia é pra você.

    Link: youtu.be/M7jR2BIuGyQ

    #debian #debian13 #lmstudio #ialocal #ibmgranite #gptoss #googlegemma #qwen #rx550 #linux

  11. New week, new slides: Run LLMs Locally

    Now including Nemotron 3 Nano Omni from Nvidia, Llama.cpp built-in tools and new slides about using Transformers.js with WebGPU for Image Recognition and OCR.

    codeberg.org/thbley/talks/raw/

    #ai #llm #llamacpp #stablediffusion #gptoss #qwen3 #glm #localai #gemma4 #nemotron #webgpu

  12. New week, new slides: Run LLMs Locally

    Now including Nemotron 3 Nano Omni from Nvidia, Llama.cpp built-in tools and new slides about using Transformers.js with WebGPU for Image Recognition and OCR.

    codeberg.org/thbley/talks/raw/

    #ai #llm #llamacpp #stablediffusion #gptoss #qwen3 #glm #localai #gemma4 #nemotron #webgpu

  13. 🔧 Fine-tunable on domain-specific data — adapts to medical, legal or enterprise environments where generic rules fail. Based on the open #gptoss model family. Available on #HuggingFace under Apache 2.0

    🚨 Caveat: #PrivacyFilter is a redaction & data minimization aid — NOT a compliance guarantee. It should be one layer in a holistic #privacybydesign approach. Always combine with human review for high-stakes use cases
    openai.com/index/introducing-o

  14. 🔧 Fine-tunable on domain-specific data — adapts to medical, legal or enterprise environments where generic rules fail. Based on the open #gptoss model family. Available on #HuggingFace under Apache 2.0

    🚨 Caveat: #PrivacyFilter is a redaction & data minimization aid — NOT a compliance guarantee. It should be one layer in a holistic #privacybydesign approach. Always combine with human review for high-stakes use cases
    openai.com/index/introducing-o

  15. New week, new update for the slides of my talk "Run LLMs Locally":

    Now including Gemma4 and Qwen3-Omni with Vision and Audio support and new slides describing Llama.cpp server parameters.

    codeberg.org/thbley/talks/raw/

    #ai #llm #llamacpp #stablediffusion #gptoss #qwen3 #glm #localai #gemma4

  16. New week, new update for the slides of my talk "Run LLMs Locally":

    Now including Gemma4 and Qwen3-Omni with Vision and Audio support and new slides describing Llama.cpp server parameters.

    codeberg.org/thbley/talks/raw/

    #ai #llm #llamacpp #stablediffusion #gptoss #qwen3 #glm #localai #gemma4

  17. New update for the slides of my talk "Run LLMs Locally": Bonsai-8B

    The latest version of Llama.cpp now supports Vulkan with 1-bit quantized models like Bonsai: 8B model having 1.1 GB in size, 2.5 GB in RAM.

    codeberg.org/thbley/talks/raw/

    #ai #llm #llamacpp #stablediffusion #gptoss #qwen3 #glm #localai

  18. New update for the slides of my talk "Run LLMs Locally": Bonsai-8B

    The latest version of Llama.cpp now supports Vulkan with 1-bit quantized models like Bonsai: 8B model having 1.1 GB in size, 2.5 GB in RAM.

    codeberg.org/thbley/talks/raw/

    #ai #llm #llamacpp #stablediffusion #gptoss #qwen3 #glm #localai

  19. New update for the slides of my talk "Run LLMs Locally": WebGPU

    Now models can run completely inside the browser using Transformers.js, Vulkan and WebGPU (slower than llama.cpp, but already usable).

    codeberg.org/thbley/talks/raw/

    #ai #llm #llamacpp #stablediffusion #gptoss #qwen3 #glm #localai #webgpu

    don't expect llm generated code to be correct ↓

  20. New update for the slides of my talk "Run LLMs Locally": WebGPU

    Now models can run completely inside the browser using Transformers.js, Vulkan and WebGPU (slower than llama.cpp, but already usable).

    codeberg.org/thbley/talks/raw/

    #ai #llm #llamacpp #stablediffusion #gptoss #qwen3 #glm #localai #webgpu

    don't expect llm generated code to be correct ↓

  21. Облачные модели Ollama в задачах code review — честное сравнение на примерах

    AI всё чаще используется в разработке: генерация кода, автодополнение, агентные IDE. Но возникает логичный вопрос - можно ли доверить LLM полноценный code review? В этой статье я решил проверить это на практике. Я сравнил несколько моделей, доступных через Ollama Cloud - Qwen 3.5, GPT-OSS и DeepSeek v3.1 - и дал им проанализировать реальные Pull Request из легаси-проекта на Python. Спойлер: некоторые модели показали неожиданно хороший результат.

    habr.com/ru/articles/1010048/

    #code_review #ollama #llm #ai_code_review #pull_request #github #open_source #deepseek #qwen #gptoss

  22. New update for the slides of my talk "Run LLMs Locally":

    Now including Reranking, Qwen 3.5 (slower than Qwen 3, but includes Vision) and loading models with Direct I/O.

    codeberg.org/thbley/talks/raw/

    #llm #llamacpp #ollama #stablediffusion #gptoss #qwen3 #glm #opencode #localai #mcp

  23. New update for the slides of my talk "Run LLMs Locally":

    Now including Reranking, Qwen 3.5 (slower than Qwen 3, but includes Vision) and loading models with Direct I/O.

    codeberg.org/thbley/talks/raw/

    #llm #llamacpp #ollama #stablediffusion #gptoss #qwen3 #glm #opencode #localai #mcp

  24. I updated the slides for my talk "Run LLMs Locally":

    Now including image generation with Qwen3 and content classification from the Qwen3Guard Technical Report paper.

    codeberg.org/thbley/talks/raw/

    #llm #llamacpp #ollama #stablediffusion #gptoss #qwen3 #glm #opencode #localai

  25. I updated the slides for my talk "Run LLMs Locally":

    Now including image generation with Qwen3 and content classification from the Qwen3Guard Technical Report paper.

    codeberg.org/thbley/talks/raw/

    #llm #llamacpp #ollama #stablediffusion #gptoss #qwen3 #glm #opencode #localai

  26. Tell ya what, #Mailroute + #Sortana / #gptoss has made my #email manageable for the first time in maybe 2 decades. Truly amazing.

  27. Tell ya what, #Mailroute + #Sortana / #gptoss has made my #email manageable for the first time in maybe 2 decades. Truly amazing.

  28. DES-Projekt Hochschulschriften: Unterstützung aus Frankreich

    Über das seit Dezember 2021 laufende DES-Projekt „Hochschulschriften“ des Vereins für Computergenealogie in Zusammenarbeit mit dem Max-Planck-Institut für Innovation und Wettbewerb in München berichtet der CompGen-Blog regelmäßig. Die Projektfamilie wächst dank einer französischen Bibliographie. Das Jahresverzeichnis der deutschen Hochschulschriften wurde zum 15. August 1885 vom preußischen Bildungsministerium ins Leben gerufen. Ein Auslöser dafür war, dass die französische […]

    compgen.de/2026/03/des-projekt

  29. I updated the slides for my talk "Run LLMs Locally":

    Now including requirements, costs, setup, llama.cpp, stable-diffusion.cpp, embeddings, function calling, opencode, image recognition, speech recognition, image generation, prompt injection and popular models like GPT-OSS, Qwen3, Qwen3-vl, Z-Image and Whisper.

    codeberg.org/thbley/talks/raw/

    #llm #llamacpp #stablediffusion #gptoss #qwen3 #opencode #php #digitalsovereignty #localai

  30. I updated the slides for my talk "Run LLMs Locally":

    Now including requirements, costs, setup, llama.cpp, stable-diffusion.cpp, embeddings, function calling, opencode, image recognition, speech recognition, image generation, prompt injection and popular models like GPT-OSS, Qwen3, Qwen3-vl, Z-Image and Whisper.

    codeberg.org/thbley/talks/raw/

    #llm #llamacpp #stablediffusion #gptoss #qwen3 #opencode #php #digitalsovereignty #localai

  31. Doing some sprucing up on my #AI #LLM powered #email filtering/tagging #Thunderbird add-on #Sortana. I had got it working with #Qwen, but realized it really didn't work for other model chat formats.

    Trying it now with #gptoss (4bit quant) and in testing it is pretty fast. Pleased so far.

  32. EdgeRunner AI just launched an offline assistant built on the open-source gpt-oss model, marking the first time open-weight LLMs are deployed with the US Army and Air Force. This could reshape how the military uses AI without internet reliance. Read more to see the implications for open-source AI and defense. #EdgeRunnerAI #gptOSS #OpenWeightModels #USMilitary

    🔗 aidailypost.com/news/edgerunne

  33. OpenAI will ChatGPT zum zukünftigen Betriebssystem machen
    OpenAI will ChatGPT auf eine neue Stufe heben und zeigt, wie Apps künftig direkt in die Chat-Oberfläche eingebettet werden. Auf der Entwicklerkonferenz in San Franc
    apfeltalk.de/magazin/news/open
    #KI #News #Apps #Betriebssystem #Canva #chatGPT #Codex #Entwickler #Entwicklerkonferenz #GPT5 #GPTOSS #KIAgenten #Monetarisierung #OpenAI #SamAltman #SDK #Spotify #Zillow