![[community profile]](https://www.dreamwidth.org/img/silk/identity/community.png){kind=small}
The Festival of Light
Aug. 28th, 2026 08:49 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
tcpipBefore I completely neglect to mention it, a few weeks ago I had the pleasure of seeing Leah Potter and friends do a little comedy show at the cosy Club Voltaire in North Melbourne. Leah makes light of some pretty delicate situations, and the show was unpretentious, friendly, and accessible. The audience seemed to be made up entirely from her old high school friends from a few decades past and friends from Darwin, which, due to a turn of events, includes me. It's rare that I attend, let alone enjoy, such events (I'm not much of a fan of "stand up" comedy), so it's to Leah's credit that I had a great night.
A few days later, I ventured out with Liza D. to Tashi Columbro's exhibit, "Ghosts", at Nightingale Gallery. Ghosts are, according to the stories, something that vanishes when the light of day shines upon them, yet when it refers to the memory of those past, they are a light that still shines in our mind long after the visceral departure, and the exhibit certainly made good use of that. To continue on the theme, shortly afterwards, Ashley M., a recent arrival in Melbourne town from central Australia, and I managed to catch the second-last evening of the Botanical Garden's "Lightscape", where the gardens become a bit of a multicoloured fairyground with a decent soundtrack.
And now, because Melbourne likes to do such things, the city is full of various light shows as part of the "Now or Never" festival, with Flinders St Station, the Town Hall, and the Yarra River bathed in colours and images. Just taking an evening promenade through the city these evenings is a rather spectacular experience. I am rather fortunate to live where I do, and I do like to take advantage of the aesthetic experience the city offers. It's either than or "la belle epoque" in Paris, and I rather missed that opportunity.
One final observation! The dual use of the English word "light" to refer to weight (physical or mental) and "light" to refer to brightness is curiously a coincidence coming from two unrelated words. It's a curiosity that light is the lightest thing of all because a photon has zero rest mass. But that's just a trick of the light (yes, I made a physics pun), because light is so light it is always moving at light speed. After all, what else could it do?
A few days later, I ventured out with Liza D. to Tashi Columbro's exhibit, "Ghosts", at Nightingale Gallery. Ghosts are, according to the stories, something that vanishes when the light of day shines upon them, yet when it refers to the memory of those past, they are a light that still shines in our mind long after the visceral departure, and the exhibit certainly made good use of that. To continue on the theme, shortly afterwards, Ashley M., a recent arrival in Melbourne town from central Australia, and I managed to catch the second-last evening of the Botanical Garden's "Lightscape", where the gardens become a bit of a multicoloured fairyground with a decent soundtrack.
And now, because Melbourne likes to do such things, the city is full of various light shows as part of the "Now or Never" festival, with Flinders St Station, the Town Hall, and the Yarra River bathed in colours and images. Just taking an evening promenade through the city these evenings is a rather spectacular experience. I am rather fortunate to live where I do, and I do like to take advantage of the aesthetic experience the city offers. It's either than or "la belle epoque" in Paris, and I rather missed that opportunity.
One final observation! The dual use of the English word "light" to refer to weight (physical or mental) and "light" to refer to brightness is curiously a coincidence coming from two unrelated words. It's a curiosity that light is the lightest thing of all because a photon has zero rest mass. But that's just a trick of the light (yes, I made a physics pun), because light is so light it is always moving at light speed. After all, what else could it do?
going through it planning funeral travel to the US
Aug. 27th, 2026 04:20 pmcimoreneMy US passport expired in 2024, and I haven't got around to renewing it yet (as mentioned in my last post) because I have a shiny new biometric Finnish passport, and didn't anticipate traveling out of the EU in the near future. (We had just started planning to meet in Montreal at WorldCon next year, but I could've used my EU passport.)
Now that my father has died suddenly the question comes up of traveling to the US. Of course, you can enter on a Finnish passport, but I was concerned about visas (Finns need one usually, US citizens do not, but if I had the wrong passport?) and also about it looking sketchy and causing discrimination of some kind at the border, given the current state of everything. I was planning to call the embassy to ask this, but I spent some time on their website first, and it gives the impression that they are very pressed for time and their definition of emergency is quite narrow. I COULD still have tried their phone queue - people doing that kind of service are usually very nice and helpful - but I was hoping to find an answer to my question on their website, which I did not. I did find that the Finnish foreign ministry is currently not recommending avoiding the US, so that seems to be a good sign. It probably means nobody with Finnish passports has been menaced, harmed, illegally interrogated, etc recently.
There is such a thing as an emergency passport if you have to travel within 2 weeks, but I might not actually need one. Would have to call them and ask about that first.
However, apparently the embassy can renew my expired passport in most likely 4-6 weeks, and I just heard from my sister that they've scheduled a service (Unitarian Universalist and non-supernatural) in a couple of months, maybe more like three, and my dad's extended family will probably be coming then. He will be cremated in a few weeks from now, so not sure why the service is that far out. My BIL's mother has offered to come from Texas to stay with them for a short while to help with cleaning and cooking. So I will talk to my mom later today, and if she or my sister prefers for me to come as soon as possible instead of later, I will attempt to do that (starting with calling the embassy to ask about travel documents). I do not care about the service itself, but I would gain something from mourning together with dad's extended family; but that's not as important a consideration as whether my presence would be helpful, or at least pleasant, for my mom and sister.
Otherwise, I should have plenty of time to have my US passport renewed through the embassy before the service. Our printer doesn't work at the moment, but I think I can print and photocopy at the library, which is a block and a half's walk.
I'm not looking forward to traveling to the US in general though. I always find it hard to sleep when traveling and my stomach seems to be upset most of the time I'm there every single time. There's universal availability of lactose-free dairy here, and I will have to either do research on the pills you can take for it or just eat mostly vegan there (but no fresh produce)!
Tips for international travel and eating on same gratefully accepted. Or knowledge about lactose pills, or whatever they are! Practical stuff. No condolences yet though please, I am not ready to wallow right now!
Now that my father has died suddenly the question comes up of traveling to the US. Of course, you can enter on a Finnish passport, but I was concerned about visas (Finns need one usually, US citizens do not, but if I had the wrong passport?) and also about it looking sketchy and causing discrimination of some kind at the border, given the current state of everything. I was planning to call the embassy to ask this, but I spent some time on their website first, and it gives the impression that they are very pressed for time and their definition of emergency is quite narrow. I COULD still have tried their phone queue - people doing that kind of service are usually very nice and helpful - but I was hoping to find an answer to my question on their website, which I did not. I did find that the Finnish foreign ministry is currently not recommending avoiding the US, so that seems to be a good sign. It probably means nobody with Finnish passports has been menaced, harmed, illegally interrogated, etc recently.
There is such a thing as an emergency passport if you have to travel within 2 weeks, but I might not actually need one. Would have to call them and ask about that first.
However, apparently the embassy can renew my expired passport in most likely 4-6 weeks, and I just heard from my sister that they've scheduled a service (Unitarian Universalist and non-supernatural) in a couple of months, maybe more like three, and my dad's extended family will probably be coming then. He will be cremated in a few weeks from now, so not sure why the service is that far out. My BIL's mother has offered to come from Texas to stay with them for a short while to help with cleaning and cooking. So I will talk to my mom later today, and if she or my sister prefers for me to come as soon as possible instead of later, I will attempt to do that (starting with calling the embassy to ask about travel documents). I do not care about the service itself, but I would gain something from mourning together with dad's extended family; but that's not as important a consideration as whether my presence would be helpful, or at least pleasant, for my mom and sister.
Otherwise, I should have plenty of time to have my US passport renewed through the embassy before the service. Our printer doesn't work at the moment, but I think I can print and photocopy at the library, which is a block and a half's walk.
I'm not looking forward to traveling to the US in general though. I always find it hard to sleep when traveling and my stomach seems to be upset most of the time I'm there every single time. There's universal availability of lactose-free dairy here, and I will have to either do research on the pills you can take for it or just eat mostly vegan there (but no fresh produce)!
Tips for international travel and eating on same gratefully accepted. Or knowledge about lactose pills, or whatever they are! Practical stuff. No condolences yet though please, I am not ready to wallow right now!
My dad died a few hours ago.
Aug. 26th, 2026 11:05 pmcimoreneHe had texted me yesterday in our conversation about antique sf. He was coherent as usual, but my sister says he was cold yesterday and tired, and those were normal signs of building infection for him. ( Read more... )
The haircutter's dilemma
Aug. 24th, 2026 01:16 pmcimoreneA part of me always wants to have my hair in a bob, but long enough to tuck behind my ears.
But another part of me always feels that my hair isn't queer enough, and urges me several times a day to cut it short, at least on the sides. (It takes forever to grow back, though, and I always want it in a bob again as soon as I cut it.)
Since I am prone to nerve pain headaches when my head gets cold behind the ears, I count this as an additional argument in favor of the bob.
But recently I'm thinking "should be more androgynous" every time I see a mirror. Ideally I should be able to switch back and forth at will.
Does this count as two wolves? I tried to get a witty title out of that, but I couldn't connect the wolves with hairstyle successfully.
But another part of me always feels that my hair isn't queer enough, and urges me several times a day to cut it short, at least on the sides. (It takes forever to grow back, though, and I always want it in a bob again as soon as I cut it.)
Since I am prone to nerve pain headaches when my head gets cold behind the ears, I count this as an additional argument in favor of the bob.
But recently I'm thinking "should be more androgynous" every time I see a mirror. Ideally I should be able to switch back and forth at will.
Does this count as two wolves? I tried to get a witty title out of that, but I couldn't connect the wolves with hairstyle successfully.
Happy Rat Birthday, Mayday and Mayhem
Aug. 23rd, 2026 09:00 pmtcpipThis week, my "animaux de compagnie", Mayday and Mayhem, turned two, which is about sixty in rat-years. With their actual birthday falling mid-week, I hosted two small parties on either side of the weekend in their well-deserved honour. Several guests paid tribute to the cult of the rat, although a special mention must go to Angela L., who, apart from providing my rodents a delicious snack made from banana, peanut butter, oats, and honey, took the opportunity to wear mouse ears to dinner. As for the feasts themselves, it was essential that ratatouille was the main dish of both events; desserts were a chocolate mousse - pronounced mouse this time, just this once - and apple tarte tartine. The ratatouille was done in traditional peasant style, as opposed to the confit byaldi (which I've made in the past) from the eponymous film. Obviously, I separated the rat-safe from rat-unsafe (garlic, onions, capsicum) ingredients for the rodent serve, and they received a generous spoonful of couscous.
Being highly social and intelligent creatures with a well-developed frontal cortex, rats have unique personalities. In his first year, Mayday was rather scittish and snatchy; as the smallest in the house, he was unsure of his place. Over the past year, he's developed into a very affectionate creature who has made a firm decision to consider humans friends. Mayhem, ever the more boisterous explorer and slightly overfond of food, has kept up with his behaviour and, like his brother, has become even closer to his human companion. I am ever-aware that they are unlikely to see a third birthday, over many years I've only had one rat, Rogue, who made it to three. Mayday has lost weight in the past few months (which is a decade in rat-time) and is a bit wobbly on his legs at times. Still, they can both zoom along for short bursts, and their tails remain high, so that is a very good prognosis for their future.
It has been said by multiple people that if reincarnation is true, they would like to come back as a rat at my place, which is quite an endorsement. Certainly, you would be guaranteed comfortable bedding, plenty of healthy food, and a lot of free-range time. I do make an effort to provide them the happiest life that their tragically short life experience brings. Apart from being creatures of profound intellectual interest (how many lectures and essays have I written about rats?) and receiving such delight (and occasional bemused annoyance) at their antics, I think that rats bring out the best in me. Celebrating their birth is indeed a great honour, and I'm so very glad to have these fascinating little creatures in my life. Here's to Mayday and Mayhem! I hope you get to have a third birthday.
Being highly social and intelligent creatures with a well-developed frontal cortex, rats have unique personalities. In his first year, Mayday was rather scittish and snatchy; as the smallest in the house, he was unsure of his place. Over the past year, he's developed into a very affectionate creature who has made a firm decision to consider humans friends. Mayhem, ever the more boisterous explorer and slightly overfond of food, has kept up with his behaviour and, like his brother, has become even closer to his human companion. I am ever-aware that they are unlikely to see a third birthday, over many years I've only had one rat, Rogue, who made it to three. Mayday has lost weight in the past few months (which is a decade in rat-time) and is a bit wobbly on his legs at times. Still, they can both zoom along for short bursts, and their tails remain high, so that is a very good prognosis for their future.
It has been said by multiple people that if reincarnation is true, they would like to come back as a rat at my place, which is quite an endorsement. Certainly, you would be guaranteed comfortable bedding, plenty of healthy food, and a lot of free-range time. I do make an effort to provide them the happiest life that their tragically short life experience brings. Apart from being creatures of profound intellectual interest (how many lectures and essays have I written about rats?) and receiving such delight (and occasional bemused annoyance) at their antics, I think that rats bring out the best in me. Celebrating their birth is indeed a great honour, and I'm so very glad to have these fascinating little creatures in my life. Here's to Mayday and Mayhem! I hope you get to have a third birthday.
среда рабочего стола, часть 2
Dec. 9th, 2011 10:34 pmuzverss
среда рабочего стола, часть 1 LJ
среда рабочего стола, часть 1 DW
среда рабочего стола часть 3 LJ
среда рабочего стола часть 3 DW
Windows
Litestep
Litestep как альтернативный рабочий стол на TS
Создание тем для LiteStep
http://litestep.info/overview/litestep-manual.html
blackbox
https://bb4win.sourceforge.net/bblean/
http://blackbox4windows.com
https://spoonm.org/bbzero/
http://www.boxshots.org/
http://www.boxshots.org/styles.zip
http://xoblite.net/
http://bb4win.sourceforge.net/styles.html
Радикальное решение всех проблем с интерфейсом Windows 8
bblean не отображается tray windows
Причина, по которой половина или более значков не отображаются, связана с тем, что explorer работает под bblean. Вам нужно будет убить его перед запуском оболочки blackbox, попробуйте что-то вроде этого
start_bb.bat
или грохнуть в taskmgr процесс explorer, затем запустить bbLean
bbIconBox.rc
Прочие
https://gnustep.github.io/windows/installer.html
http://sharpe.sourceforge.net/Download.html
https://github.com/McYoloSwagHam/win3wm
https://github.com/fuhsjr00/bug.n
https://github.com/workspacer/workspacer
https://github.com/TimUntersberger/nog
https://github.com/LGUG2Z/komorebi
https://www.astonshell.ru/
https://www.lighttek.com/talisman.htm
https://www.stardock.com/products/windowblinds/
https://www.actualtools.com/windowmanager/
https://maxto.net/ru
https://virtuawin.sourceforge.io/
https://windowspager.sourceforge.net/
https://winlaunch.org/
https://learn.microsoft.com/ru-ru/windows/powertoys/always-on-top
https://learn.microsoft.com/ru-ru/windows/powertoys/fancyzones
Ollama разворачивание и использование
Nov. 28th, 2016 03:41 pmuzverss
ознакомиться
модели
калькулятор моделей
https://ggufloader.github.io/gguf-memory-calculator.html
Размер модели | Место на диске | Оперативная память | Видеокарта --------------|----------------|--------------------|---------------------------------- 1–4B | 1–3 ГБ | 8 ГБ | Необязательна 7–8B | ~5 ГБ | 16 ГБ | Желательна, но работает и на CPU 14B | ~9 ГБ | 32 ГБ | Нужна, от 12 ГБ VRAM 30B+ | 20+ ГБ | 64 ГБ | Обязательна, от 24 ГБ VRAM характеристики моделей для установки на ноутбук | Модель | Архитектура | Вес файла | Нужно ОЗУ | Статус для ПК 16 ГБ | |-------------------|----------------------|-------------|-----------|-----------------------------------------------| | Gemma 4 E2B | Плотная (Dense) | ~1.6 ГБ | ~4–5 ГБ | Идеально 🚀 (Макс. скорость) | | Gemma 4 E4B | Плотная (Dense) | ~2.6 ГБ | ~7–8 ГБ | Отлично 🛠️ (Оптимальный умный кодер) | | Qwen 2.5 Coder 7B | Плотная (Dense) | ~4.7 ГБ | ~10 ГБ | Хорошо 📝 (Сильный кодинг, средняя скорость) | | Llama 3.1 8B | Плотная (Dense) | ~4.7 ГБ | ~10 ГБ | Хорошо 🧠 (Универсальная эрудиция) | | Gemma 4 12B | Плотная (Dense) | ~7.6 ГБ | ~12 ГБ | На грани 🐢 (Займет всю ОЗУ, будет медленной) | | Gemma 4 26B A4B | Смесь экспертов (MoE)| ~18 ГБ | ~24+ ГБ | Не запустится ❌ (Вызовет Out of Memory) |
ollama run hf.co/bartowski/Qwen2.5.1-Coder-7B-Instruct-GGUF:Q3_K_M
Распределение оперативной памяти
Суммарно процесс Ollama при генерации займет от 4.2 ГБ до 5.5 ГБ оперативной памяти. Она разделяется на две ключевые составляющие:
1. Веса модели (Фиксированная память) — ~3.6 ГБ
Базовый вес: Сама модель Qwen2.5-Coder-7B в квантовании Q3_K_M весит на диске около 3.6 ГБ.
При вызове команды Ollama полностью загрузит эти 3.6 ГБ в оперативную память. Эта цифра неизменна и не зависит от длины вашего диалога.
2. Кэш контекста (KV-Cache) (Динамическая память) — от 0.6 ГБ до 1.9 ГБ
Поскольку у вас нет видеокарты, контекст (память беседы) тоже рассчитывается в ОЗУ. Потребление памяти здесь строго зависит от того, какой размер контекста (num_ctx) задан в системе:
По умолчанию в Ollama (2048 токенов): Добавит всего около 0.6 ГБ ОЗУ. Суммарно модель займет ~4.2 ГБ.
Если выставить рабочий лимит для кода (8192 токенов): Потребует около 1.1 ГБ ОЗУ. Суммарно модель займет ~4.7 ГБ.
Если выжать максимум для 16 ГБ ОЗУ (16384 токенов): Потребует порядка 1.9 ГБ ОЗУ. Суммарно выйдет около 5.5 ГБ.
Общая карта распределения ОЗУ на ноутбуке во время генерации:
🟩 Свободно для программ: ~6.5 ГБ (под Sublime Text, терминалы и инструменты сборки).
🟨 Занято моделью (Qwen 7B Q3_K_M + KV): ~5.5 ГБ (максимальный пик при длинном коде).
🟥 Занято операционной системой: ~4.0 ГБ (стандартное потребление чистой ОС)
https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct-GGUF/tree/main
https://huggingface.co/bartowski/Qwen2.5.1-Coder-7B-Instruct-GGUF/tree/main
https://huggingface.co/bartowski/Meta-Llama-3.1-8B-Instruct-GGUF/tree/main
https://huggingface.co/bartowski/Mistral-7B-Instruct-v0.3-GGUF/tree/main
https://huggingface.co/bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF/tree/main
спецификации шаблонов, в зависимости от семейства модели для Modelfile
1. Gemma (Шаблон: Gemma 2 / 4)
2. Qwen и DeepSeek (включая DeepSeek-V3, R1 и дистилляции) (Шаблон: ChatML)
3. Meta Llama (Llama 3, 3.1, 3.2, 3.3)Использует сложную разметку с указанием ID заголовков (<|start_header_id|>) и токенами конца текста (<|eot_id|>).
4. Mistral AI (Mistral, Mixtral)Классический формат французской нейросети с тегами [INST] и [/INST]
взаимодействие с моделью через ollama
подключиться к модели
ollama run модель
например
docker exec -it ollama-server ollama run qwen2.5-coder-3b:latest
Основные команды управления сессией
/exit или /bye — полностью завершить работу и выйти из режима чата обратно в консоль вашего компьютера.
/clear — полностью очистить историю текущего диалога в памяти. Модель «забудет» всё, о чем вы говорили строчками выше, и начнет беседу с чистого листа (помогает, если контекст забился или модель начала путаться).
Команды для работы с кодом и настройками
/set — позволяет на лету менять параметры модели для текущего диалога.
Пример: /set parameter temperature 0.1 — сделает ответы модели максимально строгими, точными и техническими (идеально для кодинга).
Пример: /set parameter num_ctx 8192 — временно изменит размер окна контекста.
/show — показать подробную техническую информацию о запущенной модели.
/show info — выведет архитектуру сети, тип квантования и метаданные.
/show license — покажет лицензию модели.
/show system — отобразит текущий системный промпт (инструкцию), с которым работает модель.
Дополнительные возможности
/help — выведет на экран быструю справку по всем доступным встроенным командам.
/? — делает абсолютно то же самое, что и /help
/set parameter temperature
Available Parameters:
/set parameter seed <int> Random number seed
/set parameter num_predict <int> Max number of tokens to predict
/set parameter top_k <int> Pick from top k num of tokens
/set parameter top_p <float> Pick token based on sum of probabilities
/set parameter min_p <float> Pick token based on top token probability * min_p
/set parameter num_ctx <int> Set the context size
/set parameter temperature <float> Set creativity level
/set parameter repeat_penalty <float> How strongly to penalize repetitions
/set parameter repeat_last_n <int> Set how far back to look for repetitions
/set parameter num_gpu <int> The number of layers to send to the GPU
/set parameter stop <string> <string> ... Set the stop parameters
1. temperature <float> (Уровень креативности)
По умолчанию температура обычно равна 0.7 или 0.8. Для кодинга это много — модель начинает «фантазировать», придумывать несуществующие библиотеки или ошибаться в синтаксисе.
Рекомендация для кодинга:
Снизьте её до 0.1 или вообще до 0. Модель станет строгой, предсказуемой и будет выбирать только самые логичные и проверенные варианты написания кода.
Как ввести:
/set parameter temperature 0.1
2. repeat_penalty <float> (Штраф за повторения)
Иногда маленькие модели (вроде 3B) могут «зацикливаться» — бесконечно повторять одну и ту же строчку кода или один и тот же комментарий. Этот параметр заставляет модель подбирать новые слова, если она начинает повторяться.
Рекомендация:
Значение в районе 1.1 – 1.2 обычно идеально решает проблему циклов.
Как ввести:
/set parameter repeat_penalty 1.15
3. num_predict <int> (Максимальная длина ответа)
Этот параметр ограничивает количество токенов (слов/символов), которое модель может выдать за один раз. Если вы просите написать большой класс или развесистый скрипт, а модель внезапно обрывает ответ на середине слова — значит, она уперлась в лимит num_predict.
Рекомендация:
Для генерации больших кусков кода можно поднять лимит, например, до 2048 или 4096.
Как ввести:
/set parameter num_predict 2048
Как проверить, что параметры применились?
После того как вы ввели нужные настройки (например, изменили температуру), вы можете выполнить команду:
/show info
Когда настройки сбросятся автоматически?
При выходе из чата:
Как только вы напишете /exit или закрывающее диалог сообщение, сессия завершится. При следующем запуске ollama run параметры снова будут дефолтными.
При очистке контекста:
Если вы напишете команду /clear, модель забудет историю разговора и одновременно сбросит измененную температуру к стандартному значению.
Как зафиксировать настройки навсегда (Чтобы не вводить каждый раз)
В Modelfile для любой модели перед её сборкой через ollama create. И Ollama намертво применит эти дефолтные параметры именно к этой сборке
FROM /root/.ollama/qwen2.5-coder-3b.gguf
PARAMETER num_ctx 16384
PARAMETER temperature 0.1
PARAMETER repeat_penalty 1.1
В графических интерфейсах (Page Assist / Open WebUI)
Поскольку эти интерфейсы общаются с Ollama через единый API, ползунки настроек (Temperature, Top-P, Context Length) там работают универсально для любой выбранной модели. Они просто подставляют нужные цифры в каждый ваш запрос «на лету».
команды
ollama serve — запускает Ollama как сервис. Эта команда нужна, если вы хотите работать через API или использовать Ollama в связке с другими приложениями.
ollama create <name> — создаёт новую модель на основе существующей. Например, можно добавить свои инструкции или дообучить под конкретную задачу.
ollama show <model> — показывает подробную информацию о модели: её размер, параметры и настройки.
ollama run <model> — запускает модель в интерактивном режиме. Самая популярная команда: после её вызова можно сразу общаться с LLM прямо в терминале.
ollama stop <model> — останавливает запущенную модель, освобождая ресурсы.
ollama pull <model> — скачивает модель из репозитория. Обычно это первый шаг перед запуском: без загрузки модель не запустится.
ollama push <model> — отправляет модель в репозиторий. Полезно, если вы сделали свою сборку и хотите поделиться ей или использовать на другом устройстве.
ollama list — показывает все модели, которые установлены на вашем компьютере.
ollama ps — выводит список запущенных моделей. Удобно, чтобы посмотреть, что сейчас работает.
ollama cp <source> <destination> — копирует модель из одного места в другое, например для резервного хранения или переноса.
ollama rm <model> — удаляет модель, освобождая место на диске. Особенно актуально, учитывая, что модели могут весить десятки гигабайт.
ollama help — выводит справку по всем командам. Хорошая подсказка, если вдруг забыли синтаксис или название.
развернуть на железе
curl -fsSL https://ollama.com/install.sh | sh
или развернуть самостоятельно в ~/.local
скачать модель
файл для подключения модели
Зарегистрируйте модель в Ollama
Убедитесь, что сервер Ollama запущен (ollama serve), откройте соседний терминал и выполните команду сборки:
ollama create qwen2.5-coder:3b -f ~/Modelfile
rm ~/Modelfile
вывод моделей
ollama list
юнит systemd
mkdir -p ~/.config/systemd/user/
vim ~/.config/systemd/user/ollama.service
systemctl --user enable --now ollama.service
systemctl --user daemon-reload
systemctl --user restart ollama.service
агенты, плагины, расширения
Open WebUI (localhost:3000) — это личный локальный ChatGPT.
Aider — это автономный робот-программист. Он запускается в терминале и умеет сам создавать, изменять и удалять файлы проекта, а также делать авто-коммиты в Git.
Open Interpreter — это Исполнитель Задач. Он общается терминами операционной системы Linux. Вы можете сказать ему: «Найди в этой папке все файлы со словом "error", собери их в один архив, создай Python-скрипт для отправки этого архива и протестируй его». Он сам напишет Bash и Python скрипты и сам выполнит их по очереди внутри контейнера.
fabric (бинарь на Go) Имеет встроенные системные промты (Patterns) для написания кода, создания архитектуры и разбора логов.
Плагин llm-ollama позволяет запускать локальные языковые модели через движок Ollama прямо в терминале с автоматическим логированием истории в SQLite
плагины для браузера
Page Assist - Веб-интерфейс для локальных моделей искусственного интеллекта
https://github.com/n4ze3m/page-assist
https://chromewebstore.google.com/detail/page-assist-a-web-ui-for/jfgfiigpkhlkbnfnbobbkinehhfdhndo?hl=ru
использование fabric
https://github.com/danielmiessler/fabric
~/.local/bin/fabric --setup
Запуск: fabric -m qwen2.5-coder-3b:latest --stream
fabric -m qwen2.5-coder-3b:latest "Напиши hello world на python"
echo "Напиши hello world на python" | fabric -m qwen2.5-coder-3b:latest
cat лог_ошибок.txt | fabric -p analyze_logs (находит причину падения сервера) [v1/chat/completions].
fabric -y "ссылка_на_youtube" (cам скачивает субтитры длинного часового видео, отдаёт ИИ и выводит вам краткое саммари за 10 секунд) [v1/chat/completions].
cat плохой_код.py | fabric -p refactor_code (Переписывает код по стандартам чистого программирования) [v1/chat/completions].
Установка Aider
https://aider.chat/
https://github.com/lloydchang/Aider-AI-aider-fka-paul-gauthier-aider
на железо
необходим python 3.12 иначе баги при установке
pyenv install -s 3.12
~/.pyenv/versions/3.12.*/bin/python -m venv ~/.local/share/aider-env
source ~/.local/share/aider-env/bin/activate
# Ставим сам Aider
pip install --upgrade pip
pip install aider-chat
deactivate
ln -sf ~/.local/share/aider-env/bin/aider ~/.local/bin/aider
Запуск:
aider --model ollama/qwen2.5-coder-3b:latest
в докер
https://hub.docker.com/r/paulgauthier/aider
paulgauthier/aider 3.89GB
Собрать менее жирный
my-aider 872MB
vim Dockerfile
Соберите локальный образ
docker build -t my-aider .
Запустите my-aider
доступ к /mnt/newdisk/builds
Как вернуть модель в рабочее состояние прямо сейчас
Удаляем фейковый файл из Git
/undo
Удаляем фейковые файлы из отслеживания
/git rm -r --cached path/ hello.txt
Делаем чистый коммит изменений
/git commit -m "clean up fake files from previous session"
Вариант 1. Тестовый (Проверить создание кода и логику)
Промпт:
Напиши на Python скрипт check_system.py, который проверяет свободное место на диске, объем свободной оперативной памяти и выводит топ-5 самых тяжелых процессов в системе. Скрипт должен красиво форматировать вывод в терминал.
Используй формат diff-fenced и добавь строку 'import shutil' в самое начало файла check_system.py. Измени только импорты.
Что произойдет:
Модель создаст файл check_system.py, напишет туда рабочий код (скорее всего, используя библиотеку psutil) и сама закоммитит его в Git.
Перепиши файл check_system.py целиком. Добавь в самое начало строки:
import os
import psutil
import shutil
Вариант 2. Практический (Если вы хотите настроить линтер проекта)
Промпт:
Создай файл конфигурации .flake8 для линтера Python. Настрой его так, чтобы максимальная длина строки была 120 символов, и добавь в игнорирование папки .git, .aider_metadata и виртуальное окружение venv.
Что произойдет:
Модель создаст правильный системный конфиг для вашей папки сборок.
Вариант 3. Интерактивный (Проверить, как она исправляет свой же код)
Сначала отправьте Вариант 1, дождитесь, пока она создаст файл, а затем отправьте этот промпт вдогонку. Это покажет, как работает формат diff-fenced (когда модель меняет только нужные строчки в существующем файле).
Промпт:
Добавь в созданный скрипт check_system.py логику: если свободного места на диске осталось меньше 10%, пусть скрипт выводит предупреждение красным цветом в консоль.
если не справится
Перепиши файл check_system.py целиком. Внутри функции main() посчитай процент свободного места на диске. Если free / total < 0.10, выведи сообщение "\033[91mВНИМАНИЕ: Свободного места меньше 10%!\033[0m" красным цветом.
На что обратить внимание во время генерации:
Вы увидите, как на экране побегут строчки кода.
После завершения Aider спросит вас или сразу покажет лог вроде: Commit [хэш] feat: add system check script.
Вы сможете зайти в свою папку /mnt/newdisk/builds через обычный проводник или терминал Linux и запустить созданный файл командой python3 check_system.py.
Перепиши файл check_system.py целиком. Добавь в самое начало строки:
import os
import psutil
import shutil
Затем напиши полностью рабочий код функций check_disk_space, check_memory_usage, get_top_processes и main. Не придумывай другие файлы.
выполнить команду прямо внутри чата Aider с помощью символа ! или команды /run
!python3 check_system.py
текстовое поле Aider предназначено только для общения с ИИ. Модель не умеет сама выполнять системные команды Linux, если вы просто просите её об этом текстом
Запускайте вашей командой с флагом --edit-format whole.
Пишите модели только прямые технические задания (без «Привет» и «Как дела»).
Используйте !команда для запуска терминальных утилит (например, !python3 check_system.py).Используйте /undo для отмены, если модель 3B запуталась и сделала не то.
переключение формата
Режим вопросов
/chat-mode ask
Облегченный формат поиска и замены
/chat-mode diff-fenced
Режим полной перезаписи файлов
/chat-mode whole
выполнить команду прямо внутри чата Aider с помощью символа ! или команды /run
!python3 check_system.py
/run ls -la
Open Interpreter
Создайте Dockerfile
vim Dockerfile
добавить поддержку Си JavaScript
Соберите образ
docker build -t my-interpreter .
Запуск my-interpreter
доступ к /mnt/newdisk/builds
для 7B модели можно добавить вывод на русском
в --system_message "CRITICAL RULE: Always communicate, explain your thoughts, and reply to the user strictly in RUSSIAN language."
Промпт
Создай файл hello.txt и напиши туда "Привет из локального ИИ"
Промпт
выведи в столбик результат выполнения команды bash ls -la .
Промпт
Write and run a Python script to create a file named local_success.txt with the text "Hello from Qwen 3B!" inside.
Open WebUI
https://github.com/open-webui/open-webui
https://github.com/open-webui/open-webui/pkgs/container/open-webui
слабые модели на слабом компьютере не потянут
ghcr.io/open-webui/open-webui 5.09GB
docker stop open-webui
docker start open-webui
http://localhost:3000
загрузка
docker exec -it ollama-server ollama ps
ошибки
docker logs --tail 20 open-webui
если http://localhost:3000/error а в инкогнито открывается
Как окончательно очистить кеш этого адреса в Chrome:
Откройте вкладку с ошибкой в Chrome (http://localhost:3000).
Нажмите клавишу F12 (или Ctrl + Shift + I), чтобы открылась панель разработчика (неважно, на какой вкладке она откроется).
Теперь нажмите правой кнопкой мыши на иконку «Обновить страницу» (закругленная стрелка) в самом верху браузера, рядом с адресной строкой.
В появившемся секретном меню выберите самый нижний пункт: «Очистка кэша и аппаратная перезагрузка» (Empty Cache and Hard Reload).
После этого Chrome сотрет скрытые фоновые скрипты старого контейнера, и страница мгновенно заработает в обычном режиме. Панель F12 можно будет закрыть.
Сброс пароля
Open WebUI хранит учетные записи и чаты в базе данных SQLite, находящейся в Docker-томе по умолчанию /var/lib/docker/volumes/open-webui/_data/webui.db.
Сбросить пароль с сохранением данных можно командой
docker exec -it open-webui open-webui admin reset-password
либо удалить том, чтобы начать регистрацию заново.
взаимодействие с редакторами
Sublime Text
расширения для Sublime Text, которые взаимодействуют с моделями AI
CodeAI:
Синтаксический анализ: CodeAI выполняет синтаксический анализ и корректировку кода.
Генерация кода: CodeAI генерирует код на основе предоставленного контекста.
Обработка ошибок: CodeAI обрабатывает ошибки и предупреждения в коде.
AI Code Companion:
Автоматическое завершение: AI Code Companion автоматически завершает слова и команды.
Генерация кода: AI Code Companion генерирует код на основе предоставленного контекста.
Обработка ошибок: AI Code Companion обрабатывает ошибки и предупреждения в коде.
AI Code Generator:
Генерация кода: AI Code Generator генерирует код на основе предоставленного контекста.
Обработка ошибок: AI Code Generator обрабатывает ошибки и предупреждения в коде.
AI Code Completion:
Автоматическое завершение: AI Code Completion автоматически завершает слова и команды.
Генерация кода: AI Code Completion генерирует код на основе предоставленного контекста.
Обработка ошибок: AI Code Completion обрабатывает ошибки и предупреждения в коде.
AI Code Review:
Ревью кода: AI Code Review выполняет ревью кода и выявляет ошибки и проблемы.
Синтаксический анализ: AI Code Review выполняет синтаксический анализ и корректировку кода.
Обработка ошибок: AI Code Review обрабатывает ошибки и предупреждения в коде.
AI Code Refactoring:
Рефакторинг кода: AI Code Refactoring выполняет рефакторинг кода и улучшает его читаемость и производительность.
Синтаксический анализ: AI Code Refactoring выполняет синтаксический анализ и корректировку кода.
Обработка ошибок: AI Code Refactoring обрабатывает ошибки и предупреждения в коде.
AI Code Formatting:
Форматирование кода: AI Code Formatting выполняет форматирование кода в соответствии с правилами стиля.
Синтаксический анализ: AI Code Formatting выполняет синтаксический анализ и корректировку кода.
Обработка ошибок: AI Code Formatting обрабатывает ошибки и предупреждения в коде.
https://github.com/yaroslavyaroslav/OpenAI-sublime-text
https://github.com/dapepe/OllamaSublime
Zed
Как подключить Zed к вашей Docker-Олламе за 1 минуту:
Откройте Zed.Перейдите в его файлы глобальных настроек: нажмите комбинацию клавиш Ctrl + , (или в меню Zed -> Open Settings).
В открывшийся JSON-файл конфигурации добавьте следующий блок настроек (пропишите точное имя вашей проиндексированной модели qwen2.5-coder-3b):
~/.config/zed/settings.json
Сохраните файл настроек (Ctrl + S).
Как пользоваться ИИ прямо в коде:
Откройте любой файл с кодом проекта.Выделите кусок кода, нажмите Ctrl + Enter (вызов Inline-ассистента Zed), напишите текстовый запрос на русском (например: «Добавь обработку исключений») и нажмите Enter.
В этот момент Zed отправит сетевой запрос на порт 11434. Ваш изолированный Docker-контейнер подхватит его, прочитает веса модели с диска /mnt/newdisk/ollama_models и начнет выводить код прямо в редакторе
Vim
https://github.com/gergap/vim-ollama
https://github.com/greeschenko/vim9-ollama
https://github.com/olimorris/codecompanion.nvim
https://github.com/tzachar/cmp-ai
https://github.com/JosefAlbers/vim-lm
https://github.com/nathanbraun/nvim-ai
https://github.com/Davidyz/VectorCode
https://github.com/olimorris/codecompanion.nvim
https://github.com/Exafunction/windsurf.vim
https://github.com/whatever555/vim-code-checker
https://github.com/nomnivore/ollama.nvim
https://github.com/avante-corp/avante.nvim
https://github.com/carlos-algms/agentic.nvim
https://github.com/saya-ashen/agent-workbench.nvim
https://github.com/David-Kunz/gen.nvim
https://github.com/huggingface/llm.nvim
https://github.com/yacineMTB/dingllm.nvim
https://github.com/frankroeder/parrot.nvim
https://github.com/GeorgesAlkhouri/nvim-aider
https://github.com/nekowasabi/aider.vim
https://github.com/joshuavial/aider.nvim
https://github.com/neovim/neovim
https://github.com/LazyVim/LazyVim
развернуть в докере
для изоляции моделей лучше использовать докер
Перенос образов docker
sudo systemctl stop docker.socket
sudo mkdir -p /mnt/newdisk/docker
sudo rsync -aP /var/lib/docker/ /mnt/newdisk/docker/
sudo rsync -aP /var/lib/docker/ /mnt/newdisk/docker/
sudo vim /etc/docker/daemon.json
sudo systemctl daemon-reload
sudo systemctl start docker.socket
docker info | grep "Docker Root Dir"
docker images
sudo rm -rf /var/lib/docker
Запуск Олламы как фонового сервера (Для работы Open WebUI или Aider)
mkdir -p /mnt/newdisk/ollama_models
образ 5 гигов
Сделать более лёгкий образ
vim Dockerfile
Собрать
docker build -t ollama-debian12 .
индексация модели, запустить
rm ~/Modelfile
Чат с моделью
vim ~/.local/bin/ai-chat
chmod +x ~/.local/bin/ai-chat
Запуск сервера
vim ~/.local/bin/ai-server
chmod +x ~/.local/bin/ai-server
юнит systemd
mkdir -p ~/.config/systemd/user/
vim ~/.config/systemd/user/ollama-docker.service
systemctl --user daemon-reload
systemctl --user enable ollama-docker.service
systemctl --user start ollama-docker.service
systemctl --user stop ollama-docker.service
systemctl --user status ollama-docker.service
systemctl --user list-units --type=service --state=running
обращение к Ollama в докере
docker exec -it ollama-server ollama --help
docker exec -it ollama-server ollama list
Скачать новую модель (например, Llama 3):
curl http://127.0.0.1:11434 -H "Content-Type: application/json" -d '{"name": "llama3.1:8b"}'
или
docker exec -it ollama-server ollama run llama3
docker run -it --rm -v /mnt/newdisk/ollama_models:/root/.ollama ollama-debian12 pull llama3.1:8b
docker run -it --rm -v /mnt/newdisk/ollama_models:/root/.ollama ollama-debian12 pull gemma4:e4b
docker run -it --rm -v /mnt/newdisk/ollama_models:/root/.ollama ollama-debian12 pull qwen2.5-coder:7b
регистрация скачанных вручную в папку /mnt/newdisk/ollama_models
подробно содержание Modelfile см. в модели
vim /mnt/newdisk/ollama_models/Modelfile
FROM /root/.ollama/my_model.gguf
docker exec -it ollama-server ollama create my-custom-model -f /root/.ollama/Modelfile
vim $(pwd)/Modelfile
docker exec -i ollama-server ollama create qwen2.5-coder:7b-q4 -f - < ~/Modelfile
простейшая вебморда
vim ~/.local/bin/ai-browser
chmod +x ~/.local/bin/ai-browser
mkdir -p ~/.local/share/ai-webui
vim ~/.local/share/ai-webui/index.html
ссылки
https://github.com/ollama/ollama/
https://docs.ollama.com/linux
https://github.com/ggml-org/llama.cpp/releases
https://github.com/Aider-AI/aider
https://aider.chat/docs/install.html
https://hub.docker.com/r/paulgauthier/aider
https://github.com/lloydchang/Aider-AI-aider-fka-paul-gauthier-aider
https://github.com/open-webui/open-webui
https://github.com/danielmiessler/Fabric/releases
https://github.com/gitlawb/zero
https://github.com/taketwo/llm-ollama
https://github.com/llm-as-a-verifier/llm-as-a-verifier
https://www.reddit.com/r/ChatGPTCoding/
Ollama от А до Я: как выбрать модель, настроить и интегрировать
Оффлайн агент IDE Continue за 15 минут
https://github.com/justxor/Aiagentsfullcourse
Columbo
Aug. 21st, 2026 01:18 pmcimoreneI've started watching some of Columbo after seeing many posts about it on Tumblr. Season one is from 1971. I'm not saying the writing is bad or anything, but my favorite parts are -
- light fixtures
- everything else about the interior sets
- bell bottoms
- everything else about the clothes
- cars
- hair
- waterphone, sinister piano chords, and other soundtrack
- the stories and filmmaking techniques
NovelWriter
Aug. 20th, 2026 09:56 pmcatnessI went on yet another searching mission, looking for a text editor for creative writing, because I feel an increasing need to keep all kinds of references and drafts organized together with the project but not counted against the total word count (if it's research, it's not procrastination :P) And it has to work under Windows and Linux both, and preferably not browser-based but seamlessly jumping between several computers, synced through git (so Scrivener does not cut it).
And look what the cat dragged in! NovelWriter, a free and open-source markdown-based text editor, written in Python so fully portable, using plain text format (json, xml, md) so plays perfectly with git. Has a lot of beautiful color schemes, both light and dark, and a structure for project notes, and some special niceties like highlighting speech in quotes. So far so good, confirmed working on Ubuntu 22.04 under virtual environment, and on Windows 10. My new editor of choice :) until something shinier comes up ;)
And look what the cat dragged in! NovelWriter, a free and open-source markdown-based text editor, written in Python so fully portable, using plain text format (json, xml, md) so plays perfectly with git. Has a lot of beautiful color schemes, both light and dark, and a structure for project notes, and some special niceties like highlighting speech in quotes. So far so good, confirmed working on Ubuntu 22.04 under virtual environment, and on Windows 10. My new editor of choice :) until something shinier comes up ;)
Things
Aug. 20th, 2026 10:37 pmvassBooks
Listened to an audiobook (narrated by Rachael Beresford) combining two of Elizabeth Wein's novellas, White Eagles and Firebird. I only found out after I'd read them that they'd been written for dyslexics and 'reluctant readers'. Which explained why they seemed more simplistic and didactic than her usual. Just as angsty, though.
White Eagles is about a nineteen-year-old Polish pilot during World War II who has to fly to France with an eleven-year-old refugee. You can probably predict most of the plot twists by at least halfway through it.
Firebird is about a young Russian pilot whose loyalty to her piloting mentor puts her in violation of Stalin's Order No. 227. The mentor was such a fantastic character that she redeemed the whole book for me, despite the premise.
Still nibbling at volume 1 of Dungeon Meshi, increasingly delighted at what a freak (approving) Laiosh is.
Still reading What We Are Seeking, nothing new to report yet.
Still reading Cliff Tan's Feng Shui Modern, ditto.
Started reading Emily Nagoski's Come Together (as an audiobook narrated by the author), on maintaining the spark in long-term sexual relationships, a subject which is not relevant to me at all right now and which I can therefore approach at a safe distance. I'd been nervous about reading Nagoski despite hearing very good things about her and liking what little I'd read of her essays online, because her other two books have "for women" and "how women are different from men" all over them, and that's not something I do well with these days, for trans reasons. In fact, this book does really well at not being heterocentric or binarist or transphobic or aphobic, and manages to acknowledge that cultural ideas about masculinity and femininity exist and affect people without being essentialist about it. And the (research-based!) ideas about prioritising pleasure over desire and mapping out one's individual emotional landscape ("floor plan" in her analogy) with a view to which emotions are adjacent to each other, is genuinely interesting and I like how she presents it. I'm a bit over halfway through so far, and enjoying it a lot.
Music
I've been dipping into Archive.org's record collection.
So far I've enjoyed Musica Reservata's 1971 album of Dufay pieces, Music From The Court Of Burgundy, and Hits from Telugu Films, vol 4.
Tech
Taking a break from bothering my server to spend some time speccing out what kind of beast my next portable computing device will be. It doesn't have to be a phone, since I have an actual phone I just use for calls, messages, and data tethering. Which means it doesn't have to run Android, which is good news given the signs and portents. So it'll probably be Linux-based, but apart from that the field's pretty wide open. I'm looking at old school UMPCs and people's hand-built cyberdecks for inspiration while I decide whether I want to hack something together or buy a premade device.
Cats
Curled up together on a blanket on the floor next to my chair while I type this.
Listened to an audiobook (narrated by Rachael Beresford) combining two of Elizabeth Wein's novellas, White Eagles and Firebird. I only found out after I'd read them that they'd been written for dyslexics and 'reluctant readers'. Which explained why they seemed more simplistic and didactic than her usual. Just as angsty, though.
White Eagles is about a nineteen-year-old Polish pilot during World War II who has to fly to France with an eleven-year-old refugee. You can probably predict most of the plot twists by at least halfway through it.
Firebird is about a young Russian pilot whose loyalty to her piloting mentor puts her in violation of Stalin's Order No. 227. The mentor was such a fantastic character that she redeemed the whole book for me, despite the premise.
Still nibbling at volume 1 of Dungeon Meshi, increasingly delighted at what a freak (approving) Laiosh is.
Still reading What We Are Seeking, nothing new to report yet.
Still reading Cliff Tan's Feng Shui Modern, ditto.
Started reading Emily Nagoski's Come Together (as an audiobook narrated by the author), on maintaining the spark in long-term sexual relationships, a subject which is not relevant to me at all right now and which I can therefore approach at a safe distance. I'd been nervous about reading Nagoski despite hearing very good things about her and liking what little I'd read of her essays online, because her other two books have "for women" and "how women are different from men" all over them, and that's not something I do well with these days, for trans reasons. In fact, this book does really well at not being heterocentric or binarist or transphobic or aphobic, and manages to acknowledge that cultural ideas about masculinity and femininity exist and affect people without being essentialist about it. And the (research-based!) ideas about prioritising pleasure over desire and mapping out one's individual emotional landscape ("floor plan" in her analogy) with a view to which emotions are adjacent to each other, is genuinely interesting and I like how she presents it. I'm a bit over halfway through so far, and enjoying it a lot.
Music
I've been dipping into Archive.org's record collection.
So far I've enjoyed Musica Reservata's 1971 album of Dufay pieces, Music From The Court Of Burgundy, and Hits from Telugu Films, vol 4.
Tech
Taking a break from bothering my server to spend some time speccing out what kind of beast my next portable computing device will be. It doesn't have to be a phone, since I have an actual phone I just use for calls, messages, and data tethering. Which means it doesn't have to run Android, which is good news given the signs and portents. So it'll probably be Linux-based, but apart from that the field's pretty wide open. I'm looking at old school UMPCs and people's hand-built cyberdecks for inspiration while I decide whether I want to hack something together or buy a premade device.
Cats
Curled up together on a blanket on the floor next to my chair while I type this.
Triplet sweaters (summer 2025) roundup
Aug. 19th, 2026 01:32 pmcimoreneWe told the triplet niblings that we wanted to make them each a sweater for their 18th birthday a year and a half ago (almost). Part of this delay was a lack of strong preference, so that we had to choose what to make for two of them, but the other 8 months was just us.
Here's the finished objects:
1. from the pattern Storm Sweater (Man's) by PetiteKnit in Sandnes Garn Peer Gynt (project page). 2. from the pattern Marseille Sweater by PetiteKnit in Drops Merino Extra Fine (project page). 3. from the Drops Design vintage pattern Fisherman's Friend in Peer Gynt again (project page).
I think we'll declare this a win because it took less than a year and a half.
Here's the finished objects:
1. from the pattern Storm Sweater (Man's) by PetiteKnit in Sandnes Garn Peer Gynt (project page). 2. from the pattern Marseille Sweater by PetiteKnit in Drops Merino Extra Fine (project page). 3. from the Drops Design vintage pattern Fisherman's Friend in Peer Gynt again (project page).
I think we'll declare this a win because it took less than a year and a half.
