Meowmber of Technical staff @{redacted} before that: Silicon Architect @ainekk0, @tenstorrent, Weather @canada helps with hpc.social & @ChipsandCheese9
Ottawa && Palo Alto
- Tweets 26,047
- Following 1,057
- Followers 3,558
- Likes 112,288
Pinned Tweet
too many people who don't love The Computer have been given the privilege of working on The Computer
52
1,296
14,264
216,497
post liquidity, wonder how many will become patrons of the arts.
funding that, global access to medicine are the two main things ill be watching for
1
4
282
turns out if you hire/allocate a bunch of HFT perf nerds to design your inference stack, you get incredible results!
"Based on this framework, we have achieved high-throughput, low-latency large-scale inference services for DeepSeek V4.1 Flash. Currently, the service throughput of a single cluster is 155 trillion tokens/day, and the overall revenue is $38 per card per hour (at off-peak pricing)"
zhuanlan.zhihu.com/p/2095560…
3
5
67
1,495
🤏🤏🤏
what is this, a model for ants?
A new frontier for on-device speech.
We're open-sourcing Phonon, a suite of TTS models under 50m parameters, setting a new standard for naturalness and accuracy across languages, with the broadest inference support to date.
Phonon-31m is available under Apache 2.0.
gradium.ai/phonon🧵
2
11
412
@fclc cmp lea char retweeted
daily dose of computer architecture literature could leave you in two states
"what the hell!" - excitement, intrigued
"what the hell?" - visible confusion, disapointement
2
3
54
1,227
HPSFcon is going to one of the best food cities in north america!!
and if you're extra nice, the @Montreal account might even send you a little heart!!
You can share what you learned in 8 minutes. #HPSFcon 2027 is accepting lightning talks, along with sessions, panels, and posters. Proposals are due January 17. 🎤 Take the mic April 12-16 in Montréal!
Explore the topics and start your proposal today:
bit.ly/3VAd5FI
2
3
9
519
@fclc cmp lea char retweeted
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
1,199
3,510
30,443
3,320,737
and yet colossus was making management decisions before the concept of an "electronic computer" was a thing
the calculus of lives lost vs risk of detections is older than any active mathematicians career.
9
437
@fclc cmp lea char retweeted
>arrive at HRT party
>ask the door girl if it's quant finance or estrogen
>she doesnt understand
>pull out illustrated diagram explaining the difference between quant finance and estrogen
>she laughs and says “it's a good party sir”
>go inside
>it's an estrogen party
24
20
1,204
43,086
@fclc cmp lea char retweeted
~ and if you stare long enough into an MoE router, the router will stare back into you ~
2
3
31
2,356
ok so interesting; per Bloomberg quoting Jim, a lot of interest came after they ran DeepSeek R1 back in Q2, but those models were already a year old?
Haven't been watching if they've made progress on running more advanced models, maybe @davorVDR can comment?
3
18
1,548
model matrix hasn't been marked updated since December 2025 which is a little strange... github.com/tenstorrent/tt-me…
but they also haven't updated it to reflect since version .65, where as the latest tag is 0.8....
1
3
276
congrats to friends still at tenstorrent!
great to see them close series E, was a long time coming!
40
2,471
incomplete thought:
There's a reasonable argument to be made that cost cutting open models is actually a net benefit for AI safety in that it drops the available profit margins of most frontier model development
If that's the case, the economic hunger to reinvest at current...
2
4
35
1,240
In essence, if firms are capable of releasing (through government subsidy for geopolitical reasons) near frontier open weight models that cull the profit margins of frontier model providers, you see a natural deceleration in capital available to push frontier models forward.
...
1
8
199