罗杰斯

@JCzarlinski @__tinygrad__ Let me know what you’d like me to try, I may write a couple blog posts

@dhbrojas @__tinygrad__ was thinking of getting the p150a to tinker with myself - curious how it goes!

I will definitely regret this but the @__tinygrad__ backend is not going to write itself...

@igorjmichalak @tenstorrent Interesting, what’s the use case?

I need a @tenstorrent accelerator with this form factor and 2-4 W power draw. I have a Google Coral TPU, but I'm not too keen on TensorFlow Lite and abandoned proprietary software.

@__tinygrad__ I don’t plan on using anything from @tenstorrent beside the driver. But yeah, the first part will be documenting everything. I’ll see how much can be extracted from the TT codebases. It will probably (P=0.95) fail, but hey, it’s fun to try!

@dhbrojas Agents will not be able to do anything close to mergable. You'll end up with hacks, the tenstorrent stack isn't properly abstracted at all. The first task should be to completely document the chip to the level of the RDNA ISA manuals. This is probably a 2 year project to do right

@__tinygrad__ Do you ever bring it up to @jimkxa? What does he say?

@dhbrojas Looks like it's still missing a lot of Blackhole. It's crazy to me that $10M+ tapeouts are done without a full spec of each instruction + cycle accurate simulator.

@GaemaAI @__tinygrad__ 🫡

@dhbrojas @__tinygrad__ Good luck! From experience. 😅

@QiaoZp @__tinygrad__ 👉🏻👈🏻

@dhbrojas @__tinygrad__ 10k euros?? For personal use?

@tenstorrent Anyways, @jimkxa I beg you, take the $10M deal from @__tinygrad__. You're missing out on something great!

More seriously I think writing a software stack (runtime + compiler) for @tenstorrent is the perfect long-horizon task to evaluate coding agents on. I think I can probably spin it into a benchmark with different milestones, etc. Will report back!

@gnosisfanxyz @__tinygrad__ @tenstorrent Yeah I've been wondering the same, do you have a take @__tinygrad__?

@__tinygrad__ @tenstorrent Isn’t the TT arch meaningfully different to the multithreaded simd gpu cores TG currently targets so effectively? I think frameworks embracing the spatial dataflow possibilities of TT’s hardware is the right approach

.@tenstorrent when you are ready, we'll get you on MLPerf for $10M. Ground up stack, one 1MB pip install, zero C++20 (pure Python). From how many people I see on X trying to rewrite it, your current software approach isn't working.

Give us your money and we'll give it back to you as discounted World Cup tickets for a headline, hurray!

Mamdani's $50 World Cup tickets are distasteful. Ideological attempt at buying the sympathy of his voter base. Any marginal tax payer dollar should go to essential functions of the city, the sick or the homeless.

@willccbb Genuinely, how do you think of these?

this skill could’ve been a specialized subagent

@sumitkbehal I’m framing this tweet

never ask a woman her age, a man his salary and a value investor his semiconductor exposure

@remilouf C’est beaucoup trop. Il faut taxer les levées de fonds.

5M en IA????

@xeophon Finally we're having this conversation

Imagine SF but European (drinking alcohol before/during/after work)

separating infra and science for long context doesn't make sense, most long context science is about making computation and memory (capacity and bandwidth) feasible at scale. today's infra wouldn't support MHA on a 1T model at 1M context

@eric_alcaide @eisokant @zhijianliu_ My spot 2xB200 instance can't compete 💔

@dhbrojas @eisokant @zhijianliu_ Let's add more power to it 🔥

@zhijianliu_ fan of what you're doing at z lab! Would love to chat about getting Laguna XS.2 dflashed (is that a term 😅)