- aary
- A
- 29 Posts
- 35 Comments
localhost(2026-07-19|14:57:51) ~/rust$ rustup update stable info: syncing channel updates for stable-x86_64-unknown-linux-gnu info: latest update on 2026-07-16 for version 1.97.1 (8bab26f4f 2026-07-14) info: removing previous version of component cargo info: removing previous version of component clippy info: removing previous version of component rust-docs info: removing previous version of component rust-std info: removing previous version of component rustc info: removing previous version of component rustfmt info: downloading 6 components cargo installed 10.63 MiB clippy installed 4.71 MiB rust-docs installed 22.73 MiB rust-std installed 28.71 MiB rustc installed 77.30 MiB rustfmt installed 2.06 MiB stable-x86_64-unknown-linux-gnu updated - rustc 1.97.1 (8bab26f4f 2026-07-14) (from rustc 1.96.0 (ac68faa20 2026-05-25)) info: checking for self-update (current version: 1.29.0) localhost(2026-07-19|15:01:20) ~/rust$I think that it should do backups first and remove only in the end. Or do parallel installs.
«Примеры из книги написаны на Rust 2018 года, использующей стабильную цепочку инструментов.»
«Фрагменты кода и сообщения об ошибках создавались в конкретной версии rustc 1.70.»«Рекомендации, приведенные в книге, не раскрывают никаких аспектов функциональности async в Rust (https://oreil.ly/a9r1B), поскольку это подразумевает более продвинутые концепции и менее стабильную поддержку набора инструментов — по одному только синхронному Rust уже будет предостаточно материала.»
What you had read? Do you have a library? Do you have skills for thinking? May be you can start from applying them for yourself?
Правопреемником, взявшим на себя функции всесоюзного координационного центра, стал
Российский геронтологический научно-клинический центр (РГНКЦ) в Москве, который с 2008 года является частью РНИМУ имени Н.И. Пирогова
48GB LPDDR5 can run up to 27B dense models at INT4. MoE models like 35B-A3B (17.5GB total weights) fit comfortably.
M.2 2280 form factor allows easy integration into mini PCs, laptops, and edge devices.
обычно продается как ускоритель LQ50-24 Модуль оснащен 24 ГБ памяти LPDDR5X и обеспечивает высокую энергоэффективностьHOUMO also offers LQ50 Duo (2x M50, 320 TOPS) for 14B/32B models, and LM5070 (4x M50, 640 TOPS) for larger deployments.
https://llmdev.guide/device.html?file=houmo-lq50-48gb.md
Houmo Manjie M50
Houmo LQ50 — это компактный AI-акселератор на базе M.2 модуля от компании Houmo Intelligence, работающий на энергоэффективном чипе Manjie M50 с архитектурой совмещенных вычислений и памяти (Compute-in-Memory)
second generation Dynamic Neural Accelerator (DNA) architecture sophisticated MERA software suite, features a
- heterogeneous compiler platform, advanced quantization, and model calibration capabilities.
- MERA Model Library, with seamless interface to Hugging Face Optimum
PyTorch, TensorFlow Lite, and ONNX.
Sakyasingha Dasgupta, CEO and Founder of EdgeCortix. Founded in 2019 with R&D headquarters in Tokyo, Japan EdgeCortix is a fabless semiconductor company focused on energy-efficient AI processing of Generative-AI workloads at the edge.
https://aliexpress.ru/item/1005011631846690.html?sku_id=12000056109924620 129583 ₽ + 15892 ₽
https://discuss.huggingface.co/t/kinara-ara-2-cant-run-models/172049 «I got the Drivers and SDK from Geniatech»
«found a new Makefile on the web»
llm_model_gen.py - the script that converts models
«need a model toolchain (converter + quantizer + compiler/scheduler) that produces Ara-2 deployable artifacts»NXP’s current description of Ara Software Development Kit (ARA-SDK) explicitly frames it as:
- an extensible compiler for models including LLMs/VLMs
- “optimal execution planning” and graph scheduling
- support for quantization and multiple datatypes
Сайт NXP недоступен из России/Москвы.
Модель RM182XMC0 представляет собой готовый M.2 вычислительный модуль с 20 TOPS и объёмом памяти 2.5 ГБ (версия RK1820)
https://www.firefly.store/products/rm182xmc0-m-2-20tops-computing-module
Анонс следующего поколения (предположительно RK183X) В начале 2026 года Rockchip впервые упомянула о разработке преемника RK182X на более тонком техпроцессе (вероятно, 4 или 3 нм) с удвоенной производительностью на ватт. Ожидается в 2027 году.
По заявлению производителя, RK182X способен выполнять инференс LLM/VLM моделей объемом до 7 миллиардов параметров. Для моделей такого масштаба требуется примерно 3,5 ГБ памяти при использовании режимов INT4/FP4.
для моделей DeepSeek-R1-Distill-Qwen-7B, Qwen2.5-1.5B и Qwen2.5-3B. Выпущен специализированный SDK для LLM-инференса с поддержкой DeepSeek-R1, LLaMA-3, Qwen-2.5 в количественной версии INT4/INT8. Внедрён инструмент автоматической компиляции моделей — теперь инженеры могут переносить модели с PyTorch/TensorFlow на RK182X
Компания предлагает совместимость с популярными ИИ-фреймворками PyTorch, ONNX, TensorFlow и HuggingFace GGUF.
Theoretically .svg-s are parseable, so it is possible to prevent or remove javascript from them programmatically.
is not recommended for production systems
I think that this is because the protocol doesn’t contain means of routing (similar to “SEEN-BY” fields). But that is solvable with some coding. And current implementation will work in restricted (controlled) tree-like topologies. I think this is negotiable, and there are no too much nodes, so agreements are possible.
The main repository of this project is https://github.com/igniterealtime/Openfire License: Apache-2
Current version is v5.1.0 (2026-06-03)
The feature was implemented since v4.6.0 (2020-10-16)
«[OF-2030] - Add support for XEP-0289: Federated MUC for Constrained Environments»uninteresting notes about OF-2030
(i don’t see that link under OF-2030 - https://issues.igniterealtime.org/browse/OF-2030, and
web.archive org don’t saved it
https://download.igniterealtime.org/openfire/docs/4.6.0/changelog.html«
Loading... https://issues.igniterealtime.org/browse/OF-2030 | 17:30:40 September 20, 2022 Got an HTTP 301 response at crawl time Redirecting to... https://igniterealtime.atlassian.net/browse/OF-2030»
The link https://igniterealtime.atlassian.net/browse/OF-2030
is unavailable at web archive org https://web.archive.org/web/20220922044149/https://igniterealtime.atlassian.net/browse/OF-2030 says:426 Upgrade Required
« Похоже, на этом сайте есть проблема
https://web.archive.org/web/20220922044149if_/https://igniterealtime.atlassian.net/browse/OF-2030 вернул ошибку.
Код ошибки: 426 Upgrade Required
Проверьте, правильно ли вы ввели адрес веб-сайта.
», but available in the internet.
Contain almost no information, except that Guus der Kinderen is the author of implementation
aaryOPAtoLemmy Support@lemmy.ml•[solved] I am unable to upload a picture in my instance.
2·1 month agoHave you considered the time-verified Redis for key/value storage?
I prefer to keep the setup as it is now. But if I will need k/v storage next time, I will use something else. And then, when it will be time to upgrade, i will migrate from minio to next choice. Or, just drop this setup and create a new one.
aaryOPAtoLemmy Support@lemmy.ml•[solved] I am unable to upload a picture in my instance.Русский
2·1 month agothe problem was in reverse proxy settings. but during the reconfiguration process i also switched the setup from files to minio and from sled (pict-rs relies on an embedded key-value store called sled to store metadata) to postgres
aaryOPAtoLemmy Support@lemmy.ml•[solved] I am unable to upload a picture in my instance.Русский
2·1 month ago







« В рекомендации 1 мы разобрали, как в системе типов выражать структуры данных.» (ц) страница 24
Рекомендация 1, это часть Главы 1. И там не было разобрано, как выражать структуры данных, такие как списки, деревья, графы. Вместо этого там был описан тип enum, и как на его основе сделаны типы Option<T> и Result<TValue, TError>