Highwalker 🇦🇺

12.2K posts

Highwalker 🇦🇺 banner
Highwalker 🇦🇺

Highwalker 🇦🇺

@highwalker69

🇦🇺Never afraid of truth. I’m Austeyr-YKIYK🤷🏻‍♂️. Neurodivergent.

NSW, Australia Inscrit le Temmuz 2009
1.6K Abonnements6K Abonnés
Tweet épinglé
Highwalker 🇦🇺
Highwalker 🇦🇺@highwalker69·
AO2026 is a wrap! 🏆 We have a new King & Queen, but let’s talk about the real competition: The ones who played with grace vs. the ones who acted like total pork chops. 🥩 From Carlitos’ smile to the ‘Tunnelgate’ drama, here is the definitive Classy vs. Crassy ranking. Spoiler: The grunt finally got called. 🤫📢 x.com/highwalker69/s…
English
0
0
0
210
Highwalker 🇦🇺
Highwalker 🇦🇺@highwalker69·
@sudoingX @pupposandro I think before we pop the champagne, the key thing to note is it’s a research POC. Not denying the conceptual value, but the GH clearly states it’s diminishing returns for larger models, which is the unfortunate truth.
Highwalker 🇦🇺 tweet media
English
0
0
0
55
Sudo su
Sudo su@sudoingX·
read this carefully anon. @pupposandro wrote a single fused CUDA kernel for all 24 layers of Qwen 3.5-0.8B. one kernel launch. absolutely zero CPU round trips between layers. the result? a $900 RTX 3090 from 2020 hit 411 tok/s. apple's M5 Max hit 229. the 3090 won on speed AND efficiency 1.55x faster than llama.cpp on the same hardware. the gap between NVIDIA and Apple was never about silicon. it was software. generic frameworks waste cycles on kernel launch overhead, memory re fetches, thread synchronization. when you fuse everything into one dispatch the hardware shows what it actually has. this is the beginning of something bigger. we already proved that a 27B dense model on a single 3090 one-shots what $70K enterprise hardware cannot. now imagine what happens when someone writes kernels optimized specifically for the 3090 and the models that run best on it. not generic inference. hardware specific, model specific fused from the kernel level up. the 3090 is not a relic. it's an untapped research platform. and the people writing these kernels are proving it with data. all open source and reproducible anon.
Sandro@pupposandro

x.com/i/article/2041…

English
55
126
1.3K
98.7K
Cloudflare Developers
Cloudflare Developers@CloudflareDev·
Introducing the new /crawl endpoint - one API call and an entire site crawled. No scripts. No browser management. Just the content in HTML, Markdown, or JSON.
Cloudflare Developers tweet media
English
770
1.7K
19.8K
10.6M
Highwalker 🇦🇺
Highwalker 🇦🇺@highwalker69·
@sudoingX I came across it as well and I thought I would share this with you first. I will be testing this out as well! Appreciate all your work! Cheers mate!
English
0
0
1
33
Sudo su
Sudo su@sudoingX·
@highwalker69 haven't tried this yet. good catch. if claude code is invalidating the KV cache on every request that would explain a lot of wasted compute on local setups. going to test it on my next session and compare tok/s before and after.
English
1
0
1
209
Sudo su
Sudo su@sudoingX·
12 tok/s to 54 tok/s. same card. right model for the hardware. 5060 Ti 16GB + Qwen 3.5 9B Q4_K_XL: 54 tok/s at 262K context, thinking mode on. full model on GPU flags: -c 262144 -np 1 -fa on --cache-type-k q4_0 --cache-type-v q4_0 it's not always about squeezing the biggest model onto any card. it's about finding the right one for your hardware. i try to fit 1 trillion parameters on a single GPU because i do this for research. you don't have to. ask questions, i'll try to answer. anyone else running 16GB cards? drop your numbers and flags.
Johnny Everson@johnny_everson

@sudoingX 54tps on this setup: Qwen3.5-9B-UD-Q4_K_XL.gguf \ --ctx-size 262144 \ --temp 1.0 \ --top-p 0.95 \ --top-k 20 \ -np 1 \ -fa on \ --min-p 0.00 \ --cache-type-k q4_0 \ --cache-type-v q4_0 \ --chat-template-kwargs '{"enable_thinking":true}'

English
35
5
194
18K
Kekius Maximus
Kekius Maximus@Kekius_Sage·
What existed before the universe?
English
1.2K
37
585
97.1K
Highwalker 🇦🇺
Highwalker 🇦🇺@highwalker69·
That's such an uninformed, immature moronic take. Every country has an override clause in cases of extreme situations like wars and force majeure conditions. Ukraine specifically doesn't hold elections under martial law it is currently under. Repeating russian propaganda makes you a vatnik and a gronk wanker. ¯\_(ツ)_/¯
English
2
0
15
544
RobertFella 🇺🇸 🇺🇦
RobertFella 🇺🇸 🇺🇦@Bonkavision5·
Russia’s latest massive drone swarm just became Ukraine’s newest weapon! In January 2026, over 150 "Geran-2" (Shahed) drones were launched at Kyiv, but they didn’t all explode. Using a daring mix of GPS spoofing and Kevlar-net interceptors, Ukrainian FPV pilots like "Dimma" corralled and captured these flying bombs intact. The goal? The "crown jewels" of Russian drone tech: the MD550 engines and guidance kits. The heist was so successful that a panicked Russian command ordered an emergency self-destruct, actually firing on their own assets to prevent the theft. It didn't work. Within days, Ukraine integrated the stolen tech into their new "Beaver 2.0" drones and launched a retaliatory strike on the very base the Shaheds came from! Russia has now frozen all launches, unable to tell friend from foe in the skies. #Ukraine #UkraineWar #UkraineWarNews #RussiaIsATerroristState #RussianWarCrimes #TrumpIsARussianAsset #TrumpIsOwnedByRussia #RussiaMurdersChildren
RobertFella 🇺🇸 🇺🇦 tweet media
English
336
3.1K
11.3K
309.3K
Ukraine Battle Map
Ukraine Battle Map@ukraine_map·
Should I change by name to Iran 🇮🇷 Battle Map?
English
51
10
91
19.8K
Highwalker 🇦🇺 retweeté
Iván Ramírez de Arellano, The Jomini of the West
1/ Global Strategic Update: The global security environment is currently defined by cascading, interconnected crises that severely strain global stability and reshape the geoeconomic competition between the United States and China. This is a brief update on two of the globe's principal theaters of war, Ukraine and Iran. #UkraineWar #EpicFury #Iran #Israel
Iván Ramírez de Arellano, The Jomini of the West tweet media
English
23
583
3K
1.1M
Elon Musk
Elon Musk@elonmusk·
Gork returns today
English
11.6K
5.1K
138.9K
69.7M
Highwalker 🇦🇺
Highwalker 🇦🇺@highwalker69·
No speculation, it’s been out there for a while now. The underlying framework is mainly writing tools which manifests using Apple Intelligence for predictive text. Here are some links, if you’re a dev, you can dig deeper into this: Use predictive text on iPhone – Apple Support (AU) support.apple.com/en-au/guide/ip… Writing Tools | Apple Developer Documentation developer.apple.com/documentation/…
English
1
0
0
16
Common Woe
Common Woe@commonwoe·
@highwalker69 @andrewjclare Is your statement about ML being incorporated or the ML getting confused by old dataset in Apple's literature anywhere or is it just your speculation?
English
1
0
0
26
Andrew Clare
Andrew Clare@andrewjclare·
ignored it long enough. the iOS keyboard needs a lot, a lot of work
Andrew Clare tweet media
English
178
57
1.8K
116.6K
Highwalker 🇦🇺
Highwalker 🇦🇺@highwalker69·
@Bricktop_NAFO I’ll hang back at the Grand Hyatt (Al Khobar) while the rest potter about on the route 10 road… ¯\_(ツ)_/¯
Highwalker 🇦🇺 tweet media
English
0
0
1
63
Bricktop_NAFO
Bricktop_NAFO@Bricktop_NAFO·
This is a 265 Kilometer highway in Saudi Arabia with no bends or turns. What vehicle are you choosing?
Bricktop_NAFO tweet media
English
83
3
53
5.2K
Highwalker 🇦🇺
Highwalker 🇦🇺@highwalker69·
@commonwoe @andrewjclare Recent updates starting iOS 26 have incorporated ML to keyboard. By resetting the dictionaries, you’re letting the ML rebuild more effectively. Rather than feeding it an old and confusing dataset to train on. But, hey, you do you.
English
1
0
0
43
Common Woe
Common Woe@commonwoe·
@highwalker69 @andrewjclare This feels a lot like the same type of advice where people tell you to reinstall your OS to fix problems. I'm not resetting my training from 2007 of iOS on how i spell words to fix some occasional spelling snafus. There has to be a better way.
English
1
0
1
43
Omkar
Omkar@psomkar1·
Unpopular opinion If you can't code, you can't vibecode.
English
357
117
1.9K
90.7K