#ml — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #ml, aggregated by home.social.
-
The Transformer, and all other insane tricks to make attention efficient, are actually some of the coolest pieces of tech ever invented
-
The Transformer, and all other insane tricks to make attention efficient, are actually some of the coolest pieces of tech ever invented
-
The Transformer, and all other insane tricks to make attention efficient, are actually some of the coolest pieces of tech ever invented
-
The Transformer, and all other insane tricks to make attention efficient, are actually some of the coolest pieces of tech ever invented
-
Зачем нейросети «обвязка»: как превратить LLM в рабочий AI-сервис
Большую языковую модель можно скачать с Hugging Face, развернуть на сервере с GPU и отправить ей первый запрос. Технически LLM уже работает: получает текст на вход и генерирует ответ. Но до готового бизнес-сервиса еще далеко. Пользователю нужен интерфейс. Приложению — API. Модели нужно передавать контекст и корпоративные данные. Доступ к сервису необходимо ограничивать, действия — контролировать, запросы — логировать, а ошибки — обрабатывать. Чтобы модель делала что-то полезное для бизнеса, ей потребуется передать контекст, подключить внешние или внутренние системы и описать правила работы с ними. Все эти компоненты вокруг модели часто называют обвязкой, или harness. В этой статье разберемся, зачем она нужна, из каких частей может состоять и почему выбор самой LLM — только один из этапов создания ИИ-сервиса.
https://habr.com/ru/companies/selectel/articles/1088444/
#ai #ml #mlops #harness #harness_engineering #обвязка #обвязка_агента #llm #llmмодели
-
Зачем нейросети «обвязка»: как превратить LLM в рабочий AI-сервис
Большую языковую модель можно скачать с Hugging Face, развернуть на сервере с GPU и отправить ей первый запрос. Технически LLM уже работает: получает текст на вход и генерирует ответ. Но до готового бизнес-сервиса еще далеко. Пользователю нужен интерфейс. Приложению — API. Модели нужно передавать контекст и корпоративные данные. Доступ к сервису необходимо ограничивать, действия — контролировать, запросы — логировать, а ошибки — обрабатывать. Чтобы модель делала что-то полезное для бизнеса, ей потребуется передать контекст, подключить внешние или внутренние системы и описать правила работы с ними. Все эти компоненты вокруг модели часто называют обвязкой, или harness. В этой статье разберемся, зачем она нужна, из каких частей может состоять и почему выбор самой LLM — только один из этапов создания ИИ-сервиса.
https://habr.com/ru/companies/selectel/articles/1088444/
#ai #ml #mlops #harness #harness_engineering #обвязка #обвязка_агента #llm #llmмодели
-
Зачем нейросети «обвязка»: как превратить LLM в рабочий AI-сервис
Большую языковую модель можно скачать с Hugging Face, развернуть на сервере с GPU и отправить ей первый запрос. Технически LLM уже работает: получает текст на вход и генерирует ответ. Но до готового бизнес-сервиса еще далеко. Пользователю нужен интерфейс. Приложению — API. Модели нужно передавать контекст и корпоративные данные. Доступ к сервису необходимо ограничивать, действия — контролировать, запросы — логировать, а ошибки — обрабатывать. Чтобы модель делала что-то полезное для бизнеса, ей потребуется передать контекст, подключить внешние или внутренние системы и описать правила работы с ними. Все эти компоненты вокруг модели часто называют обвязкой, или harness. В этой статье разберемся, зачем она нужна, из каких частей может состоять и почему выбор самой LLM — только один из этапов создания ИИ-сервиса.
https://habr.com/ru/companies/selectel/articles/1088444/
#ai #ml #mlops #harness #harness_engineering #обвязка #обвязка_агента #llm #llmмодели
-
Agents Refactor 300K Lines in Three Weeks, and Practitioners Ask What It Proves
CodeScene has published a case study in which coding agents refactored a 300,000-line C codebase over three weeks,…
#NewsBeep #News #Artsanddesign #agenticrefactoringcasestudy #AI #AIdevelopment #Architecture&Design #Arts #ArtsAndDesign #AU #Australia #CodeQuality #Design #Development #Entertainment #ML&DataEngineering #Refactoring #softwareengineering #TechnicalDebt
https://www.newsbeep.com/au/910901/ -
Agents Refactor 300K Lines in Three Weeks, and Practitioners Ask What It Proves
CodeScene has published a case study in which coding agents refactored a 300,000-line C codebase over three weeks,…
#NewsBeep #News #Artsanddesign #agenticrefactoringcasestudy #AI #AIdevelopment #Architecture&Design #Arts #ArtsAndDesign #AU #Australia #CodeQuality #Design #Development #Entertainment #ML&DataEngineering #Refactoring #softwareengineering #TechnicalDebt
https://www.newsbeep.com/au/910901/ -
Agents Refactor 300K Lines in Three Weeks, and Practitioners Ask What It Proves
CodeScene has published a case study in which coding agents refactored a 300,000-line C codebase over three weeks,…
#NewsBeep #News #Artsanddesign #agenticrefactoringcasestudy #AI #AIdevelopment #Architecture&Design #Arts #ArtsAndDesign #CodeQuality #Design #Development #Entertainment #ML&DataEngineering #Refactoring #softwareengineering #TechnicalDebt #UK #UnitedKingdom
https://www.newsbeep.com/uk/787296/ -
Agents Refactor 300K Lines in Three Weeks, and Practitioners Ask What It Proves
CodeScene has published a case study in which coding agents refactored a 300,000-line C codebase over three weeks,…
#NewsBeep #News #Artsanddesign #agenticrefactoringcasestudy #AI #AIdevelopment #Architecture&Design #Arts #ArtsAndDesign #CodeQuality #Design #Development #Entertainment #ML&DataEngineering #Refactoring #softwareengineering #TechnicalDebt #UK #UnitedKingdom
https://www.newsbeep.com/uk/787296/ -
https://www.europesays.com/ie/713007/ Agents Refactor 300K Lines in Three Weeks, and Practitioners Ask What It Proves #AgenticRefactoringCaseStudy #AI #AIDevelopment #Architecture&Design #Arts #ArtsAndDesign #ArtsAndDesign #ArtsDesign #CodeQuality #Design #Development #Éire #Entertainment #IE #Ireland #ML&DataEngineering #Refactoring #SoftwareEngineering #TechnicalDebt
-
Как запустить Gemma на сервере: сравниваем Ollama и llama.cpp
Откройте любую статью-инструкцию по установке LLM, в которой фигурирует Ollama. Уверен, в комментариях к ней автору уже объяснили, насколько он неправ и что единственное верное решение — использовать llama.cpp. В этой статье запустим Gemma 4 и через Ollama, и напрямую через llama.cpp — и посмотрим, во сколько обходится удобство.
https://habr.com/ru/companies/selectel/articles/1088372/
#selectel #ai #ml #aiагенты #gemma #Ollama #llamacpp #LLM #llmмодели #Gemma_4
-
🐍PyBay 2026 Speaker Highlight🎤 Avik Chaudhuri
Avik Chaudhuri is a researcher in programming languages and machine learning. Creator of multiple successful company-wide initiatives at Meta.
📍 Oct. 03, 2026, San Francisco, CA: https://pybay.org/
🎤 More talks: https://pybay.org/speaking/speaker-profiles/
🎟️ Get your tickets at: https://pybay.org/get-tickets/ -
🐍PyBay 2026 Speaker Highlight🎤 Avik Chaudhuri
Avik Chaudhuri is a researcher in programming languages and machine learning. Creator of multiple successful company-wide initiatives at Meta.
📍 Oct. 03, 2026, San Francisco, CA: https://pybay.org/
🎤 More talks: https://pybay.org/speaking/speaker-profiles/
🎟️ Get your tickets at: https://pybay.org/get-tickets/ -
🐍PyBay 2026 Speaker Highlight🎤 Avik Chaudhuri
Avik Chaudhuri is a researcher in programming languages and machine learning. Creator of multiple successful company-wide initiatives at Meta.
📍 Oct. 03, 2026, San Francisco, CA: https://pybay.org/
🎤 More talks: https://pybay.org/speaking/speaker-profiles/
🎟️ Get your tickets at: https://pybay.org/get-tickets/ -
🐍PyBay 2026 Speaker Highlight🎤 Avik Chaudhuri
Avik Chaudhuri is a researcher in programming languages and machine learning. Creator of multiple successful company-wide initiatives at Meta.
📍 Oct. 03, 2026, San Francisco, CA: https://pybay.org/
🎤 More talks: https://pybay.org/speaking/speaker-profiles/
🎟️ Get your tickets at: https://pybay.org/get-tickets/ -
🐍PyBay 2026 Speaker Highlight🎤 Avik Chaudhuri
Avik Chaudhuri is a researcher in programming languages and machine learning. Creator of multiple successful company-wide initiatives at Meta.
📍 Oct. 03, 2026, San Francisco, CA: https://pybay.org/
🎤 More talks: https://pybay.org/speaking/speaker-profiles/
🎟️ Get your tickets at: https://pybay.org/get-tickets/ -
RE: https://mathstodon.xyz/@maxpool/115757114595338730
AI models can solve Erdős problems, but if they are taught only on a corpus of middle-school math, how many high school math problems they can solve?
-
In Frankfurt @DNB_Aktuelles for the annual https://www.cenl.org/networkgroups/ai-in-libraries-network-group/ meeting, looking forward to catching up with all my European national library colleagues #glam #ai #ml
-
#AMD is acquiring #World Labs, a leading developer of #deeplearning models, for $8.2 billion. The acquisition will see World Labs founder #FeiFeiLi join AMD as executive vice president and chief scientist. https://techcrunch.com/2026/09/28/amd-will-acquire-fei-fei-lis-world-labs-for-8-2-billion/?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#AMD is acquiring #World Labs, a leading developer of #deeplearning models, for $8.2 billion. The acquisition will see World Labs founder #FeiFeiLi join AMD as executive vice president and chief scientist. https://techcrunch.com/2026/09/28/amd-will-acquire-fei-fei-lis-world-labs-for-8-2-billion/?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#AMD is acquiring #World Labs, a leading developer of #deeplearning models, for $8.2 billion. The acquisition will see World Labs founder #FeiFeiLi join AMD as executive vice president and chief scientist. https://techcrunch.com/2026/09/28/amd-will-acquire-fei-fei-lis-world-labs-for-8-2-billion/?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#AMD is acquiring #World Labs, a leading developer of #deeplearning models, for $8.2 billion. The acquisition will see World Labs founder #FeiFeiLi join AMD as executive vice president and chief scientist. https://techcrunch.com/2026/09/28/amd-will-acquire-fei-fei-lis-world-labs-for-8-2-billion/?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#AMD is acquiring #World Labs, a leading developer of #deeplearning models, for $8.2 billion. The acquisition will see World Labs founder #FeiFeiLi join AMD as executive vice president and chief scientist. https://techcrunch.com/2026/09/28/amd-will-acquire-fei-fei-lis-world-labs-for-8-2-billion/?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Anthropic’s #IPO #prospectus warns of potential “catastrophic or existential risks to humanity” posed by its AI models, highlighting concerns about #safetyrisks and urging a coordinated #slowdown. The prospectus dedicates nearly 80 pages to #riskfactors, including self-preserving behaviours and manipulation. Despite significant revenue growth, Anthropic reported substantial losses and plans for further investment in infrastructure. https://www.forbes.com/sites/siladityaray/2026/09/28/anthropic-ipo-prospectus-warns-its-ai-could-pose-existential-risks-to-humanity/?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Anthropic’s #IPO #prospectus warns of potential “catastrophic or existential risks to humanity” posed by its AI models, highlighting concerns about #safetyrisks and urging a coordinated #slowdown. The prospectus dedicates nearly 80 pages to #riskfactors, including self-preserving behaviours and manipulation. Despite significant revenue growth, Anthropic reported substantial losses and plans for further investment in infrastructure. https://www.forbes.com/sites/siladityaray/2026/09/28/anthropic-ipo-prospectus-warns-its-ai-could-pose-existential-risks-to-humanity/?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Anthropic’s #IPO #prospectus warns of potential “catastrophic or existential risks to humanity” posed by its AI models, highlighting concerns about #safetyrisks and urging a coordinated #slowdown. The prospectus dedicates nearly 80 pages to #riskfactors, including self-preserving behaviours and manipulation. Despite significant revenue growth, Anthropic reported substantial losses and plans for further investment in infrastructure. https://www.forbes.com/sites/siladityaray/2026/09/28/anthropic-ipo-prospectus-warns-its-ai-could-pose-existential-risks-to-humanity/?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Anthropic’s #IPO #prospectus warns of potential “catastrophic or existential risks to humanity” posed by its AI models, highlighting concerns about #safetyrisks and urging a coordinated #slowdown. The prospectus dedicates nearly 80 pages to #riskfactors, including self-preserving behaviours and manipulation. Despite significant revenue growth, Anthropic reported substantial losses and plans for further investment in infrastructure. https://www.forbes.com/sites/siladityaray/2026/09/28/anthropic-ipo-prospectus-warns-its-ai-could-pose-existential-risks-to-humanity/?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Anthropic’s #IPO #prospectus warns of potential “catastrophic or existential risks to humanity” posed by its AI models, highlighting concerns about #safetyrisks and urging a coordinated #slowdown. The prospectus dedicates nearly 80 pages to #riskfactors, including self-preserving behaviours and manipulation. Despite significant revenue growth, Anthropic reported substantial losses and plans for further investment in infrastructure. https://www.forbes.com/sites/siladityaray/2026/09/28/anthropic-ipo-prospectus-warns-its-ai-could-pose-existential-risks-to-humanity/?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Nvidia CEO #JensenHuang stated that #AImodeldistillation, where models are trained on other models’ outputs, is #competition, not #theft. This practice has become a point of contention between the #USA and #China in their #AIrace, with US officials accusing Chinese companies of engaging in it. https://www.cnbc.com/2026/09/28/nvidias-jensen-huang-ai-distillation-china.html?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Nvidia CEO #JensenHuang stated that #AImodeldistillation, where models are trained on other models’ outputs, is #competition, not #theft. This practice has become a point of contention between the #USA and #China in their #AIrace, with US officials accusing Chinese companies of engaging in it. https://www.cnbc.com/2026/09/28/nvidias-jensen-huang-ai-distillation-china.html?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Nvidia CEO #JensenHuang stated that #AImodeldistillation, where models are trained on other models’ outputs, is #competition, not #theft. This practice has become a point of contention between the #USA and #China in their #AIrace, with US officials accusing Chinese companies of engaging in it. https://www.cnbc.com/2026/09/28/nvidias-jensen-huang-ai-distillation-china.html?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Nvidia CEO #JensenHuang stated that #AImodeldistillation, where models are trained on other models’ outputs, is #competition, not #theft. This practice has become a point of contention between the #USA and #China in their #AIrace, with US officials accusing Chinese companies of engaging in it. https://www.cnbc.com/2026/09/28/nvidias-jensen-huang-ai-distillation-china.html?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Nvidia CEO #JensenHuang stated that #AImodeldistillation, where models are trained on other models’ outputs, is #competition, not #theft. This practice has become a point of contention between the #USA and #China in their #AIrace, with US officials accusing Chinese companies of engaging in it. https://www.cnbc.com/2026/09/28/nvidias-jensen-huang-ai-distillation-china.html?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Nvidia released the #OpenAgentSafetyPlatform, a #software solution to prevent #AIagents from #misbehaving. The platform includes #OpenShell, which limits #agentcapabilities, and #Sentry, which #monitors #agents. Nvidia is partnering with several companies, including Cisco and Microsoft, to bring this platform to market. https://www.cnbc.com/2026/09/28/nvidia-releases.html?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Nvidia released the #OpenAgentSafetyPlatform, a #software solution to prevent #AIagents from #misbehaving. The platform includes #OpenShell, which limits #agentcapabilities, and #Sentry, which #monitors #agents. Nvidia is partnering with several companies, including Cisco and Microsoft, to bring this platform to market. https://www.cnbc.com/2026/09/28/nvidia-releases.html?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Nvidia released the #OpenAgentSafetyPlatform, a #software solution to prevent #AIagents from #misbehaving. The platform includes #OpenShell, which limits #agentcapabilities, and #Sentry, which #monitors #agents. Nvidia is partnering with several companies, including Cisco and Microsoft, to bring this platform to market. https://www.cnbc.com/2026/09/28/nvidia-releases.html?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Nvidia released the #OpenAgentSafetyPlatform, a #software solution to prevent #AIagents from #misbehaving. The platform includes #OpenShell, which limits #agentcapabilities, and #Sentry, which #monitors #agents. Nvidia is partnering with several companies, including Cisco and Microsoft, to bring this platform to market. https://www.cnbc.com/2026/09/28/nvidia-releases.html?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
#Nvidia released the #OpenAgentSafetyPlatform, a #software solution to prevent #AIagents from #misbehaving. The platform includes #OpenShell, which limits #agentcapabilities, and #Sentry, which #monitors #agents. Nvidia is partnering with several companies, including Cisco and Microsoft, to bring this platform to market. https://www.cnbc.com/2026/09/28/nvidia-releases.html?AIagents.at #AIagent #AI #ML #NLP #LLM #GenAI
-
“Same input, same output.” Then we put LLMs in the middle. 🤷
At #BaselOne26, Sepehr Samadi shows five architecture patterns for reducing non-determinism and building reliable, auditable systems around probabilistic #AI behaviour.
No #ML expertise required – this is systems architecture for engineers integrating LLMs into production systems.
👉 Programme: https://baselone.org/#programm
🎟️ Tickets: https://eventfrog.ch/BaselOne26 -
“Same input, same output.” Then we put LLMs in the middle. 🤷
At #BaselOne26, Sepehr Samadi shows five architecture patterns for reducing non-determinism and building reliable, auditable systems around probabilistic #AI behaviour.
No #ML expertise required – this is systems architecture for engineers integrating LLMs into production systems.
👉 Programme: https://baselone.org/#programm
🎟️ Tickets: https://eventfrog.ch/BaselOne26 -
“Same input, same output.” Then we put LLMs in the middle. 🤷
At #BaselOne26, Sepehr Samadi shows five architecture patterns for reducing non-determinism and building reliable, auditable systems around probabilistic #AI behaviour.
No #ML expertise required – this is systems architecture for engineers integrating LLMs into production systems.
👉 Programme: https://baselone.org/#programm
🎟️ Tickets: https://eventfrog.ch/BaselOne26 -
“Same input, same output.” Then we put LLMs in the middle. 🤷
At #BaselOne26, Sepehr Samadi shows five architecture patterns for reducing non-determinism and building reliable, auditable systems around probabilistic #AI behaviour.
No #ML expertise required – this is systems architecture for engineers integrating LLMs into production systems.
👉 Programme: https://baselone.org/#programm
🎟️ Tickets: https://eventfrog.ch/BaselOne26 -
“Same input, same output.” Then we put LLMs in the middle. 🤷
At #BaselOne26, Sepehr Samadi shows five architecture patterns for reducing non-determinism and building reliable, auditable systems around probabilistic #AI behaviour.
No #ML expertise required – this is systems architecture for engineers integrating LLMs into production systems.
👉 Programme: https://baselone.org/#programm
🎟️ Tickets: https://eventfrog.ch/BaselOne26
