Compete, @coreweave (prev @positron_ai, @googlecloud, @nvidia) Fast fleets of secure agents for all. opinions=own;
San Fierro
Joined December 2017
- Tweets453
- Following2K
- Followers776
- Likes417
Barrett Woodside retweeted
I used Opus 5.5 to help make an explainer video for @coreweave Agent Lens (agent observability + insights on prod). It's ridiculously good at these kinds of videos!!
I also used CoreWeave sandboxes to farm out the rendering.
prompts / workflow in reply
Barrett Woodside retweeted
Cognition is the first customer running @NVIDIA Vera Rubin, powered by @CoreWeave!
Hopper in 2022. Blackwell in 2024. Now Vera Rubin:
On SWE-2 inference, the new chips deliver about 4.8x more token throughput than GB200 at the same decode speed.
More compute, better agents!
Barrett Woodside retweeted
It's the final countdown. See you at #FullyConnected26 next week!
Still not registered? Last call: bit.ly/4xAeVUm
Barrett Woodside retweeted
Cool to see that you’re adding benchmarks for harness-inference co-optimization next. I wrote a bit about that here
TLDR all harnesses are basically the same because they all need to preserve the sacred KV cache hit rate of the proprietary models
BUT
what if you control the underlying infrastructure and use open models?
Then, instead of just using whatever harness you’re given, your harness and inference scheduler are aware of each other so…
There’s lots of infra optimizations that can be done :)
Session affinity is the obvious one but there’s a lot more
- speculative tool calling
- request priorities based on workload
- prefetch KV cache during tool calls
- colocate subagents and their tool calls
- prewarm tool calling sandboxes
the list goes on! I’m no expert here but know some folks that are (lmk if you’re interested in learning more)
Barrett Woodside retweeted
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
eSIMs are great and all, I really enjoy using Ubiji and/or Saily when I travel.
However, if you get your iPhone repaired and they swap the mainboard (my Lightning port stopped working), then you need to bring another device to the Apple store to re-activate. Also, turn off account security temporarily with your cell carrier.
In sum:
1. Bring a charger and charge cable to prove to the Apple store employee that the phone no longer charges
2. Turn off account migration/SIM swap security
3. Bring a second device (laptop, iPad, etc.) that's already logged into your iCloud
4. Hope for the best (like your partner on your cell plan is still in cell service and can be your second factor, if need be)
While I could restore from iCloud backup (I normally use a cable to back up my phone when I upgrade, but couldn't do that with a busted port, nor would Wi-Fi backup work without a port connection first), I was then prompted to do a 30-minute iOS upgrade (no way to bypass).
So restoring your phone after a repair, is a massive time waste with eSIMs, unless you do your homework and thoroughly prepare in advance.
OK, even if you prepare, most carriers still fail. Here's the rundown:
1. @ATT's transfer eSIM flow is broken because it requires 2FA from the phone that was repaired (via a recycled mainboard)
2. @Google_fi only lets you transfer one line and the one I wanted to move over was the secondary. I had to cancel it
3. @Visible's eSIM transfer mostly worked as intended, except I was totally logged out on the restored-from-backup device, so it prompted me to start new service rather than restore from backup. Luckily I could use e-mail as a second factor
Fun generative video clips with @krea_ai, courtesy of @ErwannMillon and team.
First pass, but should make for some fun projection mapping tomorrow night with @welovesfofficial! – at San Francisco, CA
Barrett Woodside retweeted
Quantization update!
Transformers is now compatible with models quantized with llm-compressor library from @vllm_project or models in compressed-tensors format.
This means that you can also enjoy high quality quantized models from the @neuralmagic team!
I'm so grateful for all the hard work from @trsohmers and our excellent team at @positron_ai, including the FPGA team in Liberty Lake, WA, and an amazing software team distributed across the entire country!
PSA:
If you're looking for properly quantized LLMs, huggingface.co/nm-testing (NeuralMagic, creator of Marlin) is the new Tom Jobbins huggingface.co/TheBloke
By that I mean you'll find recent models to easily run with @vllm_project, text-generation-webui, or @OpenWebUI
Barrett Woodside retweeted
#GPT4V is magical…A machine that can textually explain an image representing its own operation, and then suggest a new way to visually represent itself. What a time to be alive.
2. Windshield wipers can't be disabled while on FSD, which means sometimes they automatically activate, only to rub white road gunk into my windshield, blocking the front-facing cameras.
🧵 3/3
