the tiny corp (@__tinygrad_

the tiny corp

4,578 posts

the tiny corp

@__tinygrad__

We make tinygrad; sell tinybox for the GPU middle class. Our mission is to commoditize the petaflop.

San Diego

Joined June 2023

Replies Articles Media

Pinned
the tiny corp
@__tinygrad__
May 12
All products are now available to order! $45k for tinybox green with 5090s, and we have two variants of the tinybox pro v2, one with 5090s and one with RTX6000 Blackwells (that's a whopping 768 GB of RAM)
the tiny corp
@__tinygrad__
21h
Anthropic is the boy who cried wolf. The 1st time they cried: GPT-2 XL is too dangerous. How laughable is that now? The 2nd time they cried: Mythos is too dangerous. Now we have an open Kimi K3. Did anything happen? Next time, even if there really is a wolf, nobody will care.
the tiny corp
@__tinygrad__
22h
Started with the $39 plan and used it up, @Kimi_Moonshot is a user-aligned company I'm happy to send money to. GLM-5.2 feels Opus tier, Kimi K3 feels Fable tier!
the tiny corp
@__tinygrad__
Jul 17
As anyone who tried tinygrad would know, our value proposition has never been "the fastest kernels." We have known the whole time that very powerful AI search was coming. It's about providing a simple and understandable framework to deploy what that search finds.
JMB 🧙‍♂️
@jmbollenbacher
Jul 17
Replying to @teortaxesTex
Also a bad day for tinygrad. Their core value proposition is getting eaten by AI. Kimi can probably write better kernels than their search based compiler, and across more architectures too.
the tiny corp
@__tinygrad__
Jul 17
The time is long past to drop all export controls on NVIDIA. Some Americans are foolish enough to think they can win an AI race against China. That's unlikely. If the world switches to Chinese chips, the US will be dependent too. Let's make a global market for AI for everyone.
the tiny corp
@__tinygrad__
Jul 16
Kimi K3 is smart! I wish it was faster though, that's what I like about our local GLM-5.2. Can't wait to have Kimi at 500 tok/s. Do we know the number of active parameters?
the tiny corp reposted
Kimi.ai
@Kimi_Moonshot
Jul 16
Introducing Kimi K3: Open Frontier Intelligence 🔹 2.8 Trillion Parameters, 1 Million Context, Native Multimodal 🔹 Kimi Delta Attention enables up to 6.3x faster decoding in million-token contexts 🔹 Attention Residuals deliver ~25% higher training efficiency at <2% additional
the tiny corp
@__tinygrad__
Jul 16
Now the Chinese just need to make cheap RAM to fit it
Lisan al Gaib
@scaling01
Jul 16
Kimi K3 is 2.8T params according to their playstore app
the tiny corp
@__tinygrad__
Jul 16
AMD is trying to do some lame corporate review process for the slides of my talk. I hope it gets to happen, attempts at narrative control are so last summer. For what it's worth, I pasted my talk into ChatGPT and it thinks it's good for their brand!
the tiny corp
@__tinygrad__
Jul 15
On a car or mobile robot, you don't have the money for bandwidth or time for latency to stream your cameras to the cloud. eGPUs are going to be the top choice of smart robots everywhere. Why limit your robot to some tiny chip, stick a 5090 or 9070XT on it with tinygrad!
Mike LaBarbera
@ANSR42
Jul 15
NEWS 🧵: @comma_ai ran a livestream through the comma four, with @realGeorgeHotz driving hands free. It was connected to an AMD eGPU running unreleased v0.11.2 on a small master model. A few things I noted from the stream ↓
the tiny corp
@__tinygrad__
Jul 15
Once DDR5 prices come back to earth, some reasonably priced CPU-GPU hybrid machines for GLM 5.2 at 100 tok/s look feasible. What would you be willing to pay for an Opus level model in your living room?
the tiny corp
@__tinygrad__
Jul 15
Looks like DRAM read bandwidth on EPYC 9334 is limited by the CCD -> I/O die links, not by the RAM itself. Anyone figure out how to beat 309 GB/s?
the tiny corp
@__tinygrad__
Jul 14
tinybox green v2 blackwell ships with a monster 4xSN8100 RAID array so you don't sit around while models are loading.
the tiny corp reposted
comma
@comma_ai
Jul 13
Replying to @thdxr
All you can eat GLM tokens 🫡