Hey look, someone decided to let #AI write 55,000 lines of #Rust #code to reinvent the wheel of a 30-year-old #compression format! ๐Ÿค–๐Ÿ’พ Because who doesn't need a #sluggish, nearly-banned #RAR #clone that took only 5 weeks and an army of clanking bots to achieve? ๐Ÿ˜‚โœจ
https://bitplane.net/log/2026/05/rars/ #TechHumor #HackerNews #ngated
rars in Rust, bro

Comment la Petite Souris gรจre ses milliards de dents avec TimescaleDB

https://video.ut0pia.org/w/h37DPd3WFdgCSSZGq4imdE

Comment la Petite Souris gรจre ses milliards de dents avec TimescaleDB

PeerTube

Compressing Scrabble Dictionaries (2014)

์ด ๊ธ€์€ ์Šคํฌ๋ž˜๋ธ” ๋‹จ์–ด ๊ฒ€์ƒ‰์„ ๋น ๋ฅด๊ฒŒ ํ•˜๊ธฐ ์œ„ํ•ด ์‚ฌ์šฉํ•˜๋Š” GADDAG ์ž๋ฃŒ๊ตฌ์กฐ์˜ ๋ฉ”๋ชจ๋ฆฌ ์••์ถ• ๊ธฐ๋ฒ•์„ ์ƒ์„ธํžˆ ์„ค๋ช…ํ•œ๋‹ค. GADDAG๋Š” ๋‹จ์–ด์˜ ๋ชจ๋“  ํšŒ์ „์„ ์ €์žฅํ•˜๋Š” ํŠธ๋ฆฌ ๊ตฌ์กฐ๋กœ, ๋ฉ”๋ชจ๋ฆฌ ์‚ฌ์šฉ๋Ÿ‰์ด ๋งค์šฐ ํฌ๊ธฐ ๋•Œ๋ฌธ์— ์บ์‹œ ์ ์ค‘๋ฅ ์„ ๋†’์ด๊ธฐ ์œ„ํ•ด ์ค‘๋ณต ๋…ธ๋“œ ๋ณ‘ํ•ฉ๊ณผ ๋น„ํŠธ๋งˆ์Šคํฌ ๊ธฐ๋ฐ˜์˜ ํšจ์œจ์  ๋…ธ๋“œ ํ‘œํ˜„ ๋ฐฉ์‹์„ ์ ์šฉํ•œ๋‹ค. ํŠนํžˆ, ์ž์‹ ๋…ธ๋“œ๊ฐ€ ํ•˜๋‚˜์ธ ๊ฒฝ์šฐ๋ฅผ ์œ„ํ•œ ํŠน์ˆ˜ํ•œ ์••์ถ•๊ณผ ๋…ธ๋“œ ๊ฐ„ ์ค‘๋ณต ์ œ๊ฑฐ๋ฅผ ํ†ตํ•ด ๋Œ€๊ทœ๋ชจ ๋‹จ์–ด ๋ฆฌ์ŠคํŠธ๋„ L3 ์บ์‹œ ํฌ๊ธฐ ๋‚ด์— ์ ์žฌ ๊ฐ€๋Šฅํ•˜๋„๋ก ์••์ถ•ํ•œ๋‹ค. ์ด๋กœ ์ธํ•ด ์Šคํฌ๋ž˜๋ธ” ๋‹จ์–ด ๊ฒ€์ƒ‰ ์†๋„๊ฐ€ ๋ฉ”๋ชจ๋ฆฌ ์ ‘๊ทผ ๋ณ‘๋ชฉ ์—†์ด ํฌ๊ฒŒ ํ–ฅ์ƒ๋  ์ˆ˜ ์žˆ๋‹ค.

https://williame.github.io/post/87682811573.html

#datastructure #compression #gaddag #scrabble #cacheoptimization

Compressing Scrabble Dictionaries : Will Edward's Programming

Fc, a lossless compressor for floating-point streams

https://github.com/xtellect/fc

#HackerNews #Tech #Compression

GitHub - xtellect/fc: fc is a research-grade, lossless floating point compressor.

fc is a research-grade, lossless floating point compressor. - xtellect/fc

GitHub

ะŸั€ะธัะฝะธะปะพััŒ ะผะฝะต, ั‡ั‚ะพ ะฐะฝะดั€ะพะธะด/ะ’ะธะฒะฐะปัŒะดะธ ะฑัƒะดะตั‚ ั‚ะตะฟะตั€ัŒ ั€ะฐัะฟั€ะพัั‚ั€ะฐะฝัั‚ัŒัั ะธัะบะปัŽั‡ะธั‚ะตะปัŒะฝะพ ะฒ ัะถะฐั‚ะพะผ ะฒะธะดะต ะฒ ััƒะฟะตั€ะฟะพะฟัƒะปัั€ะฝะพะผ ั„ะพั€ะผะฐั‚ะต bzip3. ะŸั€ะพัะฝัƒะปัั ะธ ะฟะพะดัƒะผะฐะป: ะบะฐะบะฐั ะถะตัั‚ัŒ! ะะพ...ั…ะผ... - ะฟะพั‡ะตะผัƒ ะฑั‹ ะธ ะฝะตั‚?

ะ•ัะปะธ bz3 ะถะผั‘ั‚ ะปัƒั‡ัˆะต, ั‡ะตะผ ะดะฐะถะต zstd.
Vivaldi.7.9.3980.152_armeabi-v7a.apk.bz3: OK, 89177659 -> 226818751 bytes, 39.32%, 3.15 bpb

#android #vivaldiBrowser #compression #bzip3

Say, both Bob and Anne run a copy of a local, deterministic LLM. Now all Anne has to send to Bob, is a textual prompt. Allowing Bob to recreate the actual data that Anne wishes to share.

Essentially making the LLM an unprecedented compression algorithm. Anne could potentially share terabytes of high definition film, in a few thousand or less words, way under a hundred kilobytes.

What would such reproductive compression algorithms be called?

#ai #computerscience #compression

Fc, a lossless compressor for floating-point streams

fc๋Š” IEEE-754 64๋น„ํŠธ ๋ถ€๋™์†Œ์ˆ˜์  ์ŠคํŠธ๋ฆผ์„ ์œ„ํ•œ ์†์‹ค ์—†๋Š” ์••์ถ•๊ธฐ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ๋กœ, ์—ฌ๋Ÿฌ ํŠนํ™” ์ฝ”๋ฑ์„ ๋ธ”๋ก ๋‹จ์œ„๋กœ ๊ฒฝ์Ÿ์‹œ์ผœ ์ตœ์ ์˜ ์••์ถ• ๊ฒฐ๊ณผ๋ฅผ ์„ ํƒํ•œ๋‹ค. ๋ฉ€ํ‹ฐ์Šค๋ ˆ๋“œ์™€ x86-64 SIMD ๋ช…๋ น์–ด(AVX2, SSE4.2 ๋“ฑ)๋ฅผ ํ™œ์šฉํ•ด ๋น ๋ฅธ ์••์ถ• ๋ฐ ํ•ด์ œ๋ฅผ ์ง€์›ํ•˜๋ฉฐ, ํŠนํžˆ ๊ตฌ์กฐํ™”๋œ ๋ถ€๋™์†Œ์ˆ˜์  ๋ฐ์ดํ„ฐ์—์„œ ๋›ฐ์–ด๋‚œ ์••์ถ•๋ฅ ์„ ๋ณด์ธ๋‹ค. ์ธ์ฝ”๋”ฉ ์†๋„๋Š” ๋‹ค์†Œ ๋А๋ฆฌ์ง€๋งŒ ๋””์ฝ”๋”ฉ ์†๋„๊ฐ€ ๋งค์šฐ ๋น ๋ฅด๊ณ , ์‹œ๊ณ„์—ด ๋ฐ์ดํ„ฐ ์ €์žฅ์†Œ์— ์ ํ•ฉํ•˜๋‹ค. ์˜คํ”ˆ์†Œ์Šค ์—ฐ๊ตฌ์šฉ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ๋กœ Apache 2.0 ๋ผ์ด์„ ์Šค ํ•˜์— ๋ฐฐํฌ๋œ๋‹ค.

https://github.com/xtellect/fc

#floatingpoint #compression #lossless #simd #multithreading

GitHub - xtellect/fc: fc is a research-grade, lossless floating point compressor.

fc is a research-grade, lossless floating point compressor. - xtellect/fc

GitHub

Grinder12: 0.96-Bit Lossless Streaming KV-Cache (16.55x VRAM Savings

Grinder12๋Š” llama.cpp ๋Ÿฐํƒ€์ž„์˜ ํŠธ๋žœ์Šคํฌ๋จธ KV-์บ์‹œ ์••์ถ•์„ ๋ชฉํ‘œ๋กœ ํ•˜๋Š” ๋กœ์ปฌ ์ถ”๋ก  ์—”์ง„ ์—ฐ๊ตฌ ํ”„๋กœ์ ํŠธ๋กœ, 0.96๋น„ํŠธ ์œ ํšจ๊ฐ’์„ ๋‹ฌ์„ฑํ•ด FP16 ๋Œ€๋น„ 16.55๋ฐฐ VRAM ์ ˆ๊ฐ์„ ๋ณด์—ฌ์ฃผ๋Š” ์ŠคํŠธ๋ฆฌ๋ฐ ์ƒํƒœ ์ €์žฅ KV ์‚ฌ์ด๋“œ์นด ๋ฐฉ์‹์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ํ˜„์žฌ๋Š” ๋ผ์ด๋ธŒ ๋Ÿฐํƒ€์ž„ KV ๊ต์ฒด๋Š” ๊ตฌํ˜„๋˜์ง€ ์•Š์•˜์œผ๋ฉฐ, ์ œ์–ด๋œ C++ ํ™˜๊ฒฝ์—์„œ์˜ ์‹คํ—˜ ๊ฒฐ๊ณผ์™€ ๊ฐ์‚ฌ ๋กœ๊ทธ๋ฅผ ๊ณต๊ฐœํ•ด ๊ธฐ์ˆ  ๊ฒ€์ฆ๊ณผ ์ถ”๊ฐ€ ๊ฐœ๋ฐœ์„ ์œ„ํ•œ ํŒŒํŠธ๋„ˆ๋ฅผ ์ฐพ๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ๊ธฐ์ˆ ์€ ๋Œ€๊ทœ๋ชจ ์ปจํ…์ŠคํŠธ์—์„œ KV ๋ฉ”๋ชจ๋ฆฌ ์‚ฌ์šฉ๋Ÿ‰์„ ํš๊ธฐ์ ์œผ๋กœ ์ค„์ผ ์ˆ˜ ์žˆ๋Š” ๊ฐ€๋Šฅ์„ฑ์„ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค.

https://github.com/ggml-org/llama.cpp/discussions/22891

#llama.cpp #kvcache #compression #inferenceengine #streaming

Broke 1-bit KV floor (0.96-bit effective / 16.55x) with stateful streaming sidecar. Audit packet attached. ยท ggml-org llama.cpp ยท Discussion #22891

Iโ€™m an independent systems engineer operating out of Kansas through American Ironclad / ICT IronByte. Iโ€™m sharing a redacted black-box evidence packet for Grinder12, a local inference-engine resear...

GitHub