What is Hackers' Pub?

Hackers' Pub is a place for software engineers to share their knowledge and experience with each other. It's also an ActivityPub-enabled social network, so you can follow your favorite hackers in the fediverse and get their latest posts in your feed.

์ •์น˜

๊ตญ๋ฏผ์˜ํž˜ ๋Œ€๋ณ€์ธ ๊ฐ•์ œ์ถ”ํ–‰์— ์œค๋ฆฌ์œ„๊นŒ์ง€ ์†Œ๋˜์—ˆ๋Š”๋ฐ ๊ณ„์† ์ง์„ ์œ ์ง€ํ•˜๋‹ค ์˜ค๋Š˜ ์‚ฌํ‡ดํ–ˆ๋„ค.
๋‰ด์Šค๋Š” ๊ณ„์† ๋‚˜์™”๋Š”๋ฐ ์—ฌ์„ฑ๋‹จ์ฒด ๊ธฐ์‚ฌ ํ•˜๋‚˜ ์—†๋„ค
์žฅ๊ฒฝํƒœ๋•Œ๋Š” ๋ฐœ๋น ๋ฅด๊ฒŒ ๋ฐœํ‘œํ•˜๋”๋‹ˆ?
0

ใจใ„ใ†ใ“ใจใงใ€ ใซ
ไปŠๅ›žใฏใ€Œๅฑ•็คบๆž ใ€ใซใฆๅ‡บๅฑ•ใ•ใ›ใฆใ„ใŸใ ใใพใ™๏ผ๏ผ๏ผ

็ด ๆ™ดใ‚‰ใ—ใ„ๆ–นใ€…ใซๅ›ฒใพใ‚ŒใฆไปŠๅนดใ‚‚ๅ‚ๅŠ ใงใใ‚‹ใ“ใจใŒๅ…‰ๆ „ใงใ™
ใ•ใ‚‰ใชใ‚‹ไฝœๅ“ใ‚’ใŠ่ฆ‹ใ›ใงใใ‚‹ใ‚ˆใ†ๅŠชใ‚ใฆใพใ„ใ‚Šใพใ™ใฎใง
ใชใซใจใž๏ผใ‚ˆใ‚ใ—ใใŠ้ก˜ใ„ใ„ใŸใ—ใพใ™๏ผ

ๅๅ‰ๆ€ฅใซๅค‰ใˆใกใ‚ƒใฃใฆใ”ใ‚ใ‚“ใชใ•ใ„ๅ็พฉใฏใ€ŒใŸใณใ™ใ‚‹ใ€ใงใ„ใใพใ™

3

12์›” 8์ผ
1๋ฒˆ์งธ ์‹œ๋„์— ์„ฑ๊ณตํ–ˆ์Šต๋‹ˆ๋‹ค. ์ดˆ์„ฑ์ด ใ……ใ……์ธ ๋‹จ์–ด๋ฅผ ๋งž์ถœ ์ˆ˜ ์žˆ์„๊นŒ์š”?

https://gyeon.postica.app

๊ฒฌ๋ผ์–ํ˜ธ์šฐ

1
0
1
0
0
2
0
16
0

์ด์   ๊ฐœ์ธ์ด ๊ทน๋ณตํ•  ์ˆ˜ ์—†๋Š” ๊ฒƒ๋“ค์ด ๋„ˆ๋ฌด ๋งŽ์•„์ง€๋Š” ์‹œ๋Œ€์ด๋‹ค๋ณด๋‹ˆ,

์–ด๋–ป๊ฒŒ ๊ทน๋ณตํ•˜๋А๋ƒ๊ฐ€ ์•„๋‹ˆ๋ผ ์–ด๋–ป๊ฒŒ ์งŠ์–ด์ง€๊ณ  ์‚ฌ๋А๋ƒ์— ๋Œ€ํ•ด ์ƒ๊ฐํ•ด๋ณผ๋•Œ์ธ๊ฑฐ๊ฐ™์Šต๋‹ˆ๋‹ค.

0

ใ‚นใƒžใƒ›ใฎ่จญๅฎš่ฆ‹็›ดใ—ใฆใŸใ‚‰ๅฎŒๅ…จใซๅ‡บ้…ใ‚Œใฆใ—ใพใฃใŸ
ๆ–ฐใ—ใ„ใ‚นใƒžใƒ›ใซใชใฃใฆใ‹ใ‚‰ใ‚ใฃใกใ‚ƒๅฟซ้ฉใงใ™
ๅ‰ๆฉŸ็จฎใ‹ใ‚‰ใƒกใƒขใƒช2ๅ€ใ€CPUๅ‡ฆ็†่ƒฝๅŠ›3ๅ€ใใ‚‰ใ„ใซใชใฃใฆใ‚‹ใฎใงใใ‚Šใ‚ƒใ‚ตใ‚ฏใ‚ตใ‚ฏใงใ™ใ‚ˆใญ
โ€‹:kumapu_zanzou:โ€‹โ€‹:maku_kaisoku:โ€‹

1
2
1
0

่บซ็‚บไธ€ๅ€‹ไธๅฅฝ็›ธ่™•็š„ไบบ๏ผŒ็œ‹ๅˆฐ้‚ฃ็จฎใ€Œ90% ็š„ไบบ้ƒฝ็”จ้Œฏไบ†๏ผใ€็š„ๆจ™้กŒ๏ผŒ้ƒฝๆœƒๆƒณๅซไฝœ่€…ๆŠŠ่ณ‡ๆ–™ไพ†ๆบๅๅ‡บไพ†ใ€‚

0

hits you over the head with a german precision manufactured KraftfahrzeughaftpflichtversicherungsmaklerbรผroklingelschildbeleuchtungsdrahtabisolierzangenhandgrifffarbeninspektionsrobotergelenkschmierfettstoรŸpressendruckluftkartuschenkompressormotorkommutatorkohlebรผrstenpresshรผlsenstanzwerkzeughรคrteรถlrezepturgeheimsafeschlรผsseletuinahtsternzwirnwickelmaschinenspindeldrehbankwerkzeughalterungsschraubenkopfschablonenkonstruktionszeichnungsdรผnnlinienbleistiftspitzerklingenschraubenkopfschlitzsรคgeblattzahnschleifmaschinenstromversorungskabelisolierungsmixerrรผhrstab

0
1
0
0

็พๅœจ Android ่ทŸ Linux ไธŠๆœ‰ๅ“ชไบ› Mastodon ๅฎขๆˆถ็ซฏๆ”ฏๆดๅผ•็”จๅ•Š๏ผŸTusky ๅซๆˆ‘่ฏ็ตก้–‹็™ผๅœ˜้šŠ๏ผŒTokodon ็›ดๆŽฅๅœจ้€š็Ÿฅๆฌ„่ฎŠๅ‡บไธ€ๅ€‹้ป‘่‰ฒๅคง็ชŸ็ชฟ wwww

0
0
0
0
0

ๅ‹•็‰ฉๅคฉ็”Ÿ็†ฑๆ„›่‡ช็”ฑ๏ผŒไฝ†ๅพŒๅคฉ็’ฐๅขƒๅฏไปฅๆ”น่ฎŠ้€™ๅ€‹ๅคฉๆ€งใ€‚

ไธ€้šปๅพžๅฐๅฐฑ่ขซๅ›š็ฆ็š„้ณฅ๏ผŒๅณไพฟ็ตฆไป–่‡ช็”ฑ๏ผŒไป–ไนŸไธ่ฆ‹ๅพ—ๆœƒ้ฃ›่ตฐ๏ผŒๅ› ็‚บ่‡ช็”ฑๆ„ๅ‘ณ่‘—่ฆ็‚บ่‡ชๅทฑ็š„ๅฎ‰ๅ…จ๏ผŒ็‚บ่‡ชๅทฑ็š„ๆบซ้ฃฝ่ฒ ่ฒฌใ€‚

่ซ‹AI็•ซ็š„ๅœ–ใ€‚็พๅœจAI็•ซ็š„ๅœ–้ƒฝๅฏไปฅ็Žฉใ€Œๅคงๅฎถไพ†ๆ‰พ็ขดใ€ ๏ผŒ็œŸๆœ‰่ถฃ๏ฝž
0
1
0

่บซ็‚บไธ€ๅ€‹ไธๅฅฝ็›ธ่™•็š„ไบบ๏ผŒ็œ‹ๅˆฐ้‚ฃ็จฎใ€Œ90% ็š„ไบบ้ƒฝ็”จ้Œฏไบ†๏ผใ€็š„ๆจ™้กŒ๏ผŒ้ƒฝๆœƒๆƒณๅซไฝœ่€…ๆŠŠ่ณ‡ๆ–™ไพ†ๆบๅๅ‡บไพ†ใ€‚

0

feel like there's a bit of nuance lost when people argue about usability in Linux. Just because I know how to do something by editing a file in a very specific directory that changes between distros does not mean that I think it's good ui/ux. I can be frustrated with how something should be a simple GUI toggle vs rawdogging dconf. I can also still be smart; it does not make me dumb to want shortcuts.

it is genuinely infuriating how many conversations regarding this start off with "are you too stupid to use vim to edit a file in ~/.config/?". No, I do it all the time. I don't think that everyone should have to fucking do that for basic functionality, though.

0

ใ€Œ่™Ž่€ณๅฅณใซใชใฃใŸใจใฆใ€ใฎใŠใพใ‘ใ‚คใƒฉใ‚นใƒˆใŒๆ›ดๆ–ฐใ ใž
ใชใ‚“ใ‹ใƒˆใƒƒใƒ—ใƒšใƒผใ‚ธใซไฟบใฎๅก—ใฃใŸ่ฆšใˆใฎใชใ„ใ‚ซใƒฉใƒผใ‚ณใƒžใŒๅ‡บใฆใ‚‹ใžใ€่‰ใ ใž
https://share.comic-fuz.com/manga/3899

2

feel like there's a bit of nuance lost when people argue about usability in Linux. Just because I know how to do something by editing a file in a very specific directory that changes between distros does not mean that I think it's good ui/ux. I can be frustrated with how something should be a simple GUI toggle vs rawdogging dconf. I can also still be smart; it does not make me dumb to want shortcuts.

0
1
0
1

ๅฏ่ƒฝๆ˜ฏๅฒไธŠๅฎฃๅ‚ณๆŠ•ๅ…ฅๆœ€ๅคš็š„้ธ่ˆ‰
็ต‚ๆ–ผ้”ๆˆ็›ฎ็š„ไบ†

ๅฎŒๅ–„ๅˆถๅบฆๅพŒๆŠ•็ฅจ็އๆœ€้ซ˜็š„ไธ€ๆฌก

่จป๏ผšๅŒๆ™‚ๆœฌๅฑ†้ธๆฐ‘ๆฏ”ไธŠๅฑ†ๅฐ‘ไบ†33่ฌใ€‚็ธฑ่ง€้Žๅพ€้ธ่ˆ‰๏ผŒๆฐ‘ไธปๆดพๅœจ้ธๆฐ‘้–“็š„ๆ”ฏๆŒๅบฆๅคงๆฆ‚ๆ˜ฏ60%๏ผŒๆ›ๅฅ่ฉฑ่ชช๏ผŒๅŸบๆœฌไธŠๆฐ‘ไธปๆดพๆ”ฏๆŒ่€…ไพ่ˆŠๆฒ’ๆœ‰ๅƒ่ˆ‡้€™ๆฌก้ธ่ˆ‰

0
0
0

Some highlights from !

Keynote - Imogen Wright - How Complex Systems Taught Me To Fail
youtube.com/live/MObVZKZr5vY

Sofia Toro - How to teach your language to Python (with CPython!)
youtube.com/watch?v=JhFKjiEWHWA

Meagen Voss - Building more accessible Python-powered websites
youtube.com/watch?v=KrtUTEZzD6U

Panel - From Contributor to Founder: Turning Python Projects into Products
Carol Willing, Inessa Pawson, Deborah Hanus, Leah Wasser
youtube.com/live/NB2Q9dbLwVc

0

็พๅœจ Android ่ทŸ Linux ไธŠๆœ‰ๅ“ชไบ› Mastodon ๅฎขๆˆถ็ซฏๆ”ฏๆดๅผ•็”จๅ•Š๏ผŸTusky ๅซๆˆ‘่ฏ็ตก้–‹็™ผๅœ˜้šŠ๏ผŒTokodon ็›ดๆŽฅๅœจ้€š็Ÿฅๆฌ„่ฎŠๅ‡บไธ€ๅ€‹้ป‘่‰ฒๅคง็ชŸ็ชฟ wwww

0
0
0
0
1
0
1
1
0
0
2

์–ด์ œ ์ž๊ธฐ ์ „์—, ๊ทธ๋ƒฅ ํฅ๋ฏธ ์œ„์ฃผ๋กœ ์บ๋…ผ RF ์นด๋ฉ”๋ผ๋“ค์˜ ์งˆ๊ฐ ํ‘œํ˜„ ์•ฝ์ ์„ ๋ฉ”๊พธ๋Š” RAW ํ˜„์ƒ๋ฒ•์„ ๋ฌผ์–ด๋ดค๊ณ , ์ œ๋ฏธ๋‚˜์ด๋Š” ์ˆ˜๋งŽ์€ ์„ค์ •๊ฐ’๋“ค์„ ์Ÿ์•„๋ƒˆ๋‹ค.

... LLM์€ ๋ฌด์„œ์›Œ.; ๋ฌผ๋ก  ํ‹€๋ฆฌ๊ณ  ์†Œ์šฉ์—†๋Š” ์ •๋ณด๋“ค๋„ ์žˆ์—ˆ์ง€๋งŒ ๋ฐ˜์€ ๊ฑด์ง„ ๋“ฏ.;; Dual pixel AF์˜ ๋‚ด๋ถ€ ๋ฏธ์„ธ ํšŒ์ ˆ๊ณผ AAํ•„ํ„ฐ์˜ ๋ธ”๋Ÿฌ๋ฅผ ์นด์šดํ„ฐ ์น˜๋Š” ๋ฐฉ๋ฒ•์„ ๋งŒ๋“ค์–ด๋‘๋‹ค๋‹ˆ Rawtherapee ๊ฐœ๋ฐœํŒ€๋“ค๋„ ์ •์ƒ?์€ ์•„๋ƒ. ๐Ÿคฃ

์‚ฌ์ง„์€ ๋ฏธ์„ธ ๋””ํ…Œ์ผ์„ ์ฃฝ์–ด๋ผ ๋Œ์–ด์˜ฌ๋ ค๋ณธ ์Šต์ž‘. ๋ฌผ๋ก  ๋ณด์ •ํ•œ๋‹ค๊ณ  ํƒ€์‚ฌ ์„ผ์„œ๋งŒํผ ์ข‹์•„์ง€๋Š” ๊ฑด ์•„๋‹ˆ์ง€๋งŒ ์•ฝ์ ์€ ๋งŽ์ด ๋ณด์™„๋˜๋”๋ผ.

0

Zebra-Llama โ€“ Towards efficient hybrid models

Link: arxiv.org/abs/2505.17272
Discussion: news.ycombinator.com/item?id=4

arXiv logo

Zebra-Llama: Towards Extremely Efficient Hybrid Models

With the growing demand for deploying large language models (LLMs) across diverse applications, improving their inference efficiency is crucial for sustainable and democratized access. However, retraining LLMs to meet new user-specific requirements is prohibitively expensive and environmentally unsustainable. In this work, we propose a practical and scalable alternative: composing efficient hybrid language models from existing pre-trained models. Our approach, Zebra-Llama, introduces a family of 1B, 3B, and 8B hybrid models by combining State Space Models (SSMs) and Multi-head Latent Attention (MLA) layers, using a refined initialization and post-training pipeline to efficiently transfer knowledge from pre-trained Transformers. Zebra-Llama achieves Transformer-level accuracy with near-SSM efficiency using only 7-11B training tokens (compared to trillions of tokens required for pre-training) and an 8B teacher. Moreover, Zebra-Llama dramatically reduces KV cache size -down to 3.9%, 2%, and 2.73% of the original for the 1B, 3B, and 8B variants, respectively-while preserving 100%, 100%, and >97% of average zero-shot performance on LM Harness tasks. Compared to models like MambaInLLaMA, X-EcoMLA, Minitron, and Llamba, Zebra-Llama consistently delivers competitive or superior accuracy while using significantly fewer tokens, smaller teachers, and vastly reduced KV cache memory. Notably, Zebra-Llama-8B surpasses Minitron-8B in few-shot accuracy by 7% while using 8x fewer training tokens, over 12x smaller KV cache, and a smaller teacher (8B vs. 15B). It also achieves 2.6x-3.8x higher throughput (tokens/s) than MambaInLlama up to a 32k context length. We will release code and model checkpoints upon acceptance.

arxiv.org ยท arXiv.org

0

ํ•ด๋ฐ€ํ„ด์ด ํŽ˜๋ผ๋ฆฌ์—์„œ ์ฃฝ์‘ค๋Š”๊ฑฐ ์ƒ๊ฐํ•˜๋ฉด ์˜ฌ ํ•œํ•ด ์žซ ๊ฒฌ๋Ž ๋‹ค. ๋‚ด๋…„์—” ์ด์ œ ๋‚œ์žฅํŒ๋ ํ…Œ๋‹ˆ ํ•œ๋ฒˆ ๋ด…์‹œ๋‹ค :)

0