Small machines. Big possibilities.

Joined April 2026
Include

Only show posts containing:

Exclude

Hide posts containing:

Time range
-
Minimum likes
The last week I have been running a comparison of SGLang vs TensorFold vs vLLM on different variants of GLM 5.3-Flash. I wanted to see if tensor fold speed gave up any material performance. TB2.1 is the last run and should be done this weekend. Excited to release the full results.
1
5
81
Qwen3.8-Flash-Next and GLM-5.3-Flash, optimized for local AI. ⚡️ NVIDIA’s FP4-Local checkpoints reduce memory requirements, giving you more room for agents and longer context on DGX Spark and RTX PRO 6000 Blackwell systems. Explore the NVIDIA Local AI Optimized Checkpoints collection: huggingface.co/collections/n… @Alibaba_Qwen @Zai_org
29
58
11
499
37,432
Haven’t even tried it yet.
21
Great to see @elonmusk take the lead here. Hopefully others will follow. xitter.cf/secscottbessent/status…
Replying to @jackclarkSF
It would be interesting to donate it directly to Trump Accounts for the children. Directly goes to kids. No administration drain. It could also be a model for other companies. At a $2T valuation that would be $400B to seed the reinvention of Social Security. Only would need a handful of other large AI and tech companies to do the same to match what is in Social Security today with larger growth opportunities. I think there is no better way to guarantee the public benefits by the companies success.
13
Replying to @DavidSacks
Why not just put it directly into children’s Trump Accounts? Wouldn’t that be the best public good?
3
356
Replying to @jackclarkSF
It would be interesting to donate it directly to Trump Accounts for the children. Directly goes to kids. No administration drain. It could also be a model for other companies. At a $2T valuation that would be $400B to seed the reinvention of Social Security. Only would need a handful of other large AI and tech companies to do the same to match what is in Social Security today with larger growth opportunities. I think there is no better way to guarantee the public benefits by the companies success.
2
2
1
91
1,597
Replying to @0xSero
Will be trying this! Thanks!
210
Eye spy some RAM in my black cars center console. Do I completely ruin the 45min drive to the airport for my wife by asking about it?
17
swapped vllm for tensorfold on my two DGX Sparks and it literally doubled same glm 5.3 flash weights, nothing else touched > 24 → 50 tok/s single > 40 → 85 tok/s 4 streams > automationbench performed better too @MiaAI_lab @ashxhart you two cooked 🔥
7
2
37
3,365
localsonly retweeted
As I said many times before, TensorFold is a real turning point for local AI. Just try it for yourself, we have a recipe for everyone! All of my TensorFold recipes can be found here: mia-ai.net/models?for=engine…
TensorFold by @ashxhart is one of our newest Inception startups and has been making waves in the local AI community. It's an open-source inference engine for running models on your own hardware. Finally got around to experiencing it for myself. I'm running Qwen3.8 Flash Next in NVFP4 on it across my two DGX Sparks, and built a paper reader to put its concurrency to work. It splits an arXiv paper into its sections and reads them all at once, one request per section, then merges the results into a one-page brief tagged back to each source section. Per-request and combined tok/s update live as it runs.
19
13
2
212
18,479
2.1x faster than GLM non tensor fold. Very excited to see local and the DGX keep improving. We will see if there is any performance difference. Thanks @MiaAI_lab for the work.
27
Replying to @realEstateTrent
Think Harmon was Brown also up there on the all time list. And hotel + Cornell.
5
1,468
Replying to @realEstateTrent
Congrats! Making the right decision for your family is how a home should be made, not investment timing. We bought in FL mid 2024. Not ideal timing, probably top of the FL market, but the deal made sense for what we are trying to do with our family. Also doesn’t make sense, paying it down as quickly as possible. About 45% now paid off.
441
He’s such a good actor.
Ben Affleck reveals he writes Python, understands convolutional neural networks, worked extensively with GPUs, and used his celebrity status to get private looks at Google and OpenAI’s video models “I’ve always been kind of into computers since I was young. Then, when film started to move from analog film to digital, I became more interested in that aspect of it. The visual-effects workflow for many years has included machine learning, so I can write pretty shitty Python scripts and stuff like that. “With convolutional neural networks, which were the precursors to what the transformer can do, which is much more computation simultaneously, you would do things like look at what’s called a tensor. That’s the numerical translation of a visual image in numbers, like the batch number, the frame number and the red, green and blue values of each pixel in each frame. It’s just that simple. That numeric is called a tensor. “You’d use a convolutional neural network to identify patterns that reveal what’s called edge detection or feature extraction, which is identifying patterns well enough to know, this is where the window ledge is, so we can more easily take the green-screen image out and replace it with something. “That was familiar to me early on because, prior to Artists Equity, I had a small visual-effects company. I’ve worked with GPUs a lot too. The visual-effects guys said, ‘Hey, you should see. There are a couple: Google and this other company, OpenAI, are doing really interesting stuff with transformers in video.’ “I’ve learned that I can actually just call up and go, ‘Hey, it’s Ben Affleck. Can I come see what you’re doing?’ Sometimes people say yes, to my astonishment.”
15
Artificial Analysis has been given access by Reflection and is independently benchmarking Beam Early indicators suggest Beam will be one of the most token-efficient open models we've seen for its level of intelligence. Congratulations @reflection_ai on the announcement!
Introducing Beam: a highly efficient agentic open model with 501B total parameters and 23B active. - Frontier reasoning efficiency - Advances the Western open frontier on coding & agentic tasks - Trained end-to-end from scratch Full weights release this month. Learn more about Beam: reflection.ai/beam
51
78
16
925
77,550
Replying to @MiaAI_lab
Thank god my gut told me to pull the trigger on the 2nd DGX Spark in late August.
2
304
GLM 5.3 on 2 DGX Sparks has been pretty impressive so far. Not only more accurate than Qwen 3.8 Flash-Next and DeepSeek V4, but so much faster. About 20-30% improvement compared to 3.8 and double on DSV4. Exited to see some of the speed improvements from TensorFold over the next few weeks. Running on actual workflows - this is the first local model I have run that would actually work as an overnight worker.
1
80
Replying to @sudoingX @TechMDAI
Didn’t think I would get much lift from the 2nd DGX but GLM 5.3 is really good. Can’t wait for some of the speed improvements too.
1
104
I got it on my account. They do expire by end of year unfortunately. Would have been nice to be able to spread them out over 1 year.
21