Drew Botwinick retweeted
sparkrun 0.4.0 is released. this is a major update with a lot of changes! sparkrun.dev/getting-started…
also
sparkrun run @experimental/deepseek-v4.1-flash-1017-b12x-dspark-sglang-tp4
for a ds4.1f 1017 recipe that is based on Siv's work for ds41f.
RT @spark_arena: 🟢 Spark Arena is going all green!
Eugene joined NVIDIA back in May, and now Raphael and Drew are joining too!
All three…
Drew Botwinick retweeted
Great initiative by @nvidia
Today, NVIDIA is committing $1 billion over the next five years to help U.S. scientists advance research in quantum computing, healthcare, and energy.
This builds on more than two decades of work with U.S. national labs.
Read the release ➡️ nvda.ws/3VAeXhQ
Drew Botwinick retweeted
Need some help with your Spark(s)?
Want to learn more about the NVIDIA ecosystem?
Want a community with highly technical and helpful people and no jerks?
Join our Discord Server:
discord.gg/GH5kRgv6ZD
Drew Botwinick retweeted
Check out Raphael Amorim's Mia-AiLab/GLM-5.3-Flash-EXL3-TR3-4bpw on 2-nodes NVIDIA DGX Spark cluster with vLLM!
You can also run it locally with:
sparkrun update
sparkrun registry trust experimental
sparkrun run @experimental/glm-5.3-flash-exl3-dflash2-2x-vllm
View full benchmark at spark-arena.com/benchmark/10…
Drew Botwinick retweeted
GLM scoring close to Spark frontier on the tok/s per GB per node.
@Zai_org
spark-arena.com/leaderboard?…
I just wanted to say that this is still a problem more than two years later. If @Chime cares this little about resolving ID theft, I wouldn't want to know what they think of their customers. They tell you what you want to hear on phone but never follow through.
@Chime Never tried to open account, was a victim of ID theft. Called. Was told acct closed and would be handled. I disputed the entry on my credit report, you blocked me from making changes. Emails also say acct closed, been dealing with this for YEARS! WTF. DO NOT TRUST CHIME.
I want to send a thank you to @raphafacts! none of this exists without you!
Our first thank-you goes to Drew Botwinick (@DrewBotwinick) and Eugene Rakhmatulin (@eugr).
They’re special contributors not only because they’re friends and founding members of Spark Arena, but because they’ve dedicated so much time and energy to making the Spark community stronger. Through projects like Sparkrun.dev, spark-vllm-docker, and llama-benchy, as well as their invaluable contributions to the NVIDIA Developer Forums, they’ve helped make experimentation more accessible, practical, and collaborative for everyone.
Without you, none of this would have happened.
Drew Botwinick retweeted
🎂 Spark Arena is 6 months old!
Six months ago, we introduced Spark Arena to the NVIDIA DGX Spark community with a simple belief
We wanted to turn experiments scattered across forum threads into something reproducible, searchable, and useful to everyone. From real results, on real hardware, by the community.
Six months later, the part we’re most proud of isn’t the platform.
It’s the people who showed up and built it with us. ❤️
People who ran benchmarks.
People who spent nights debugging broken builds.
People who tested new models and runtimes.
People who shared configurations instead of keeping them to themselves.
People who found bugs, challenged results, contributed code, helped others, and pushed their Sparks far beyond what we thought possible.
So for our 6-month anniversary, we want to celebrate them.
Over the next few days, we’ll honor some of the contributors who helped turn Spark Arena from an idea into a real community.
🏆 6 months.
⚡ Hundreds of experiments.
🌎 One incredible community.
And we’re only getting started.
Happy 6 months, Spark Arena! 🚀
Drew Botwinick retweeted
In this article @DrewBotwinick shows the work with Qwen3.8-27B on sparkrun.dev for standardized recipes.
Same benchmark profile. Same workflow:
launch → health check → benchmark → teardown → results. View article
Drew Botwinick retweeted
Weights are out and it's time to code!
Experimental MiniMax M3 for DGX Spark is out now!
- 4x sparks:
sparkrun run @experimental/minimax-m3-v0-nvfp4-4x
- 2x sparks:
sparkrun run @experimental/minimax-m3-v0-nvfp4-2x-reap50
The REAP50 is even more experimental, but testing shows that it's coherent in responses.
View Spark Arena benchmark at spark-arena.com/benchmark/ef…
NVIDIA Dev Forum thread:
forums.developer.nvidia.com/…
