@skalskip92 Real craftmanship and kudos to @PINTO03091.
English
Diego Garcia Rieckhof
86 posts
I did a lot of manual image labeling over the years, but this is insane
Super PINTO@PINTO03091
@skalskip92 You might not believe it, but I simply manually annotated over 2,000,000 human body parts with ultra-precise detail. Probably no one else could do that.
English
@TheEcomNomad @NVIDIAAI Agree! Data is always the hard part, good thing about this is that it can help me to speed up the annotation process finding the moments im looking for, creating the segments and getting a first draft of the analyses I want
English
Meet Alcides, a tiny-box AI agent for UFC fight analysis.
It identifies key moments, does grounding and fighter segmentation, and turns fight footage into tactical summaries.
Built with DGX Spark (@NVIDIAAI), Cosmos-Reason2, SAM 3.1, and Llama 3 70B.
English
Been running Cosmos Reasoning 2 8B NVFP4 on my DGX Spark (@NVIDIAAI). Getting ~900 tok/s prefill and ~40 tok/s decode with vLLM, no optimization yet.
Enjoying it for temporal + spatial reasoning. Curious if anyone has pushed it further or found better configs on a DGX.
English
What are you guys using to SSH into your desktop from your phone nowadays?
English
true bravery is when you don't even read the plan claude generates
kache@yacineMTB
it takes a lot of discipline to not commit code you haven't read
English
@brainyface @pimma77 @thsottiaux Felt the same, my account resets today haha better late than never haha
English
@pimma77 @dgarieck @thsottiaux I know, right? ... I was up late last night coding and woke up early today, fingers crossed that I'd see the reset - but nope. Of course my account is scheduled to reset this afternoon anyway. Bummer - a bunch of us that have business seats lost a couple days worth of Codex.
English
Three million people are now using Codex weekly - up from two million a little under a month ago. Incredible to see the growth. Thank you to all of you and to the ecosystem we’re part of. To celebrate, we’re resetting rate limits so you can keep building, and we’ll reset them every additional 1M users until we reach 10M, so we can keep celebrating along the way.
Enjoy and thank you!
English
GIF
English
English
@thsottiaux Much appreciated, but disappointingly I am not seeing the reset. Is this rolling out? Usually I see the reset right after your post.
English
@KiyanSo Common sense but social media lives from being either black or white, doesn’t understand about grey scales. I need to check the data again but the midfield is a problem too, I think that overall the team is pretty unbalanced with or without mbappe
English
Good analysis. You have to be very careful reading too much into on-off data -- not because you can't value the analysis but because you'd need to see a list of all those opponents and take it into context. Was he rested in an easier game? Was it an open game or a low block? Who else was on the field? This is not basketball where you can extrapoliate on-off numbers in a real way. There are 11 players on the field. You need at least a 1000 minute sample size to measure anything and even then, you have to look at the context fully.
I'm neither advocating that Real Madrid are better or worse with / without Mbappe. But there are so many variables at play.
Don Amancio@DonShelby__
Veo muchos sacando conclusiones sobre los resultados del Madrid con y sin Mbappé que ha subido Ramón (no los he confirmado). Hay que tener cuidado con eso. Por ejemplo, excluyendo partidos contra equipos fuera de primera en Copa, el Madrid en 461 partidos en 10 temporadas con Toni Kroos en el campo ganó el 64,4% (297 victorias) y perdió el 16,1% (74 derrotas), ganó el 64,0% (270 victorias) y perdió el 15,9% (67 derrotas) cuando el alemán fue titular y ganó el 79,3% (69 victorias) y perdió solo el 12,6% (11 derrotas) cuando el alemán no jugó. ¿Eso quiere decir que el Madrid era mejor equipo sin Kroos en el campo durante esas 10 temporadas?
English
@NielsRogge @openclaw Might help you spark-arena.com/leaderboard, besides other model deployment recipes you can check different settings and speeds
English
Wtf... Qwen3.5-35B-A3B took 3 minutes (!!) to answer my simple question, "What's on my calendar today?" via @openclaw
I don't know what these local LLM fellas are running on, but a DGX Spark sure is not the best thing
English
@rzschernsdorf @Teknium @mr_r0b0t @LottoLabs @sudoingX @MiniMax_AI @Dell @TheAhmadOsman Agree with it, at the end of the day depends on what’s the final use case and the ROI.
English
@dgarieck @Teknium @mr_r0b0t @LottoLabs @sudoingX @MiniMax_AI @Dell @TheAhmadOsman Its even 3x if we compaire Asus Ascent GX10 vs RTX6000. And yeah of course you also need a computer for the GPU.
I think for agentic workloads, experimentation and exploration it might still be the better decision, considering how much token demand it creates.
English
Dear algorithm, please gift @LottoLabs @sudoingX @Teknium and @mr_r0b0t one of these so we can run Hermes with local post-trained @MiniMax_AI M.7! We promise to make some really amazing promotional content!
@Dell
Lotto@LottoLabs
Maybe we should try qwen 3.5 397b at home
English
@rzschernsdorf @Teknium @mr_r0b0t @LottoLabs @sudoingX @MiniMax_AI @Dell @TheAhmadOsman 2x only considering the gpu card. We have to add ram, motherboard, etc. also is more power intensive. It’s obvious that rtx 6000 will be faster than DGX but it’s not true that it’s only twice the price
English
@Teknium @mr_r0b0t @LottoLabs @sudoingX @MiniMax_AI @Dell I think @TheAhmadOsman will have more real data on this. But the bandwidths in a spark are limited. Its amazing for its size, but its capped by its size.
RTX6000 are 2x the cost for roughly 5x the performance according to benchmarks by others: github.com/casualcomputer…
English
@ahmadierfan999 @skalskip92 @NVIDIAAIDev @PLcostes Wasn’t available in trackers when I tried it, so had to implement and test it before using it on my videos and it did better than the ones available at least for this use case
English
Third round of experiments. Rider isolation works, but tracking is still challenging due to camera motion and wave dynamics. Surfing is very different from most sports.
Using my DGX Spark (@NVIDIAAIDev) and planning to test the new Trackers 2.3.0 update by @skalskip92.
English
English