Yea a bit of parallel tasks and subagents with those tasks, but I've had a similar workflow in the past with GLM 5.2/3/F. This model does a better job of working til completion, but NOT over working. It's correctly balanced this time.

fall recovery is definitely one of the more fun RL policies that I've worked on. still a ways to go, this one would definitely break a lot of stuff as is, but we'll continue improving! still just a lil baby model.

I kneel... training @asimovinc to transfer from lying on the ground (back/belly/side) to a half kneel, with hopes of translating next from a half kneel to full stand. Trying to find the best way to get up off the ground.

you can buy a whole dell workstation with an RTX Pro 6K WS inside of it for less than most places will sell you just the pro 6k

dario begged me in dms to not do agent swarms at home but whatever, this is nemotron 120B A12B running 64 concurrents at a bit over 100 tok/sec each for an aggregate of almost 7K tok/sec on a single GB300 (Asus DGX Station) believe it or not, the agents did NOT magically just start hacking the planet tho. weird.

wym? maybe I am misunderstanding you but they did include Opus 5.5 there.

Trying to train a fall recovery policy for the @asimovinc humanoid Getting up is hard!

Nemotron 3.5 Super VL (120B A12B) is now the leading US open source model. Congrats to the NVIDIA AI team! I am still doing testing of this model, but I think the ranking would be even stronger if we were just measuring agentic coding performance. This one may very well be a solid choice for a speedy local coding agent model I'm running many things in parallel, will get proper average and max numbers in time, but just randomly prodding this model for some quick speed checks and over 550 tok/sec. All vanilla settings. Once optimized 600+, maybe even 650+, will be trivial, and so far it does appear to be a token efficient problem solver. Really exciting to see what looks like a solid release so far from nvidia!

It is another good morning!

reminds me of when i kept my server rack in the kitchen

It's a good morning btw

Robot used here is @sourccey from Vulcan Robotics. If youre interested in an actually useful robot to R&D on like this, definitely hit them up. For a deeper dive into how all this is working at the moment: piped.video/watch?v=dGkJe3aI…

No training, no sim2real gaps, just a prompt and a dream. The problem in robotics today is every task is a complex combination of locomotion, manipulation, vision, planning, and a bunch of other things that fall apart once you diverge from training data or characteristics from sim. This is why you see so many demos with perfect lighting conditions and always the same, or precisely placed objects, specifically chosen for the task and demo. You never see these things deployed in the real world despite seeing demos that look convincing, because everything falls flat in the real world with real unseen conditions. Every single deployment is a unique edgecase. Here, it's just a multimodal LLM (DSV4.1F) doing a task it was never deliberately trained for, using cameras it's never seen, servos/motors it didn't learn on, in an environment it's never dealt with and a form factor never tested. It is currently slow, this video is 10x speed, but I am very confident it can be sped up as is, but also this is basically the slowest this level of intelligence will ever be again. The more I work with general purpose multimodal LLMs, the more I am realizing that ACT, VLAs, and all this other stuff is probably just a sideshow and a distraction, and it's just going to largely get eaten up by general purpose models that just happen to also do robotics. I don't think it'll always be LLMs, but the fact that LLMs can do some of the most challenging robotics tasks tells me it's kind of over for a large class of robotics training. This does not include RL, however, for low level things, like gait, reaching, and other capabilities or "skills" for hardware. RL is still a king here and you really do want these models to be super fast and super small. RL is very good here and safe I'd say. Decision models like Clef, Jev...etc remain mostly useless imho for robotics. I look forward to someone proving out a useful recipe that works these things into robotics, but I don't currently see it.

That's ~ 63K tok/sec prefill and obviously 450 tok/sec decode is also lovely

I cannot escape the connection of the Reflection 70B fiasco with Reflection the company. Every time I see this name I get irked a little bit. Maybe it's just me but I wonder how many people remember the totally unrelated Reflection 70B model and associate those feelings consciously or subconsciously and it hurts this co's rep. How much of that negativity exists in LLMs too I wonder?

PRETTTTY sure yall just playing runescape

yeah go 100-lane pcie 5.0 switch from c-payne 100%. He's sold out atm i just checked, but you can email him any questions you have and he'll get back to you super fast usually and can prob tell you when he'll restock. pcie switch diagram i vibed a while back:

i heard you liked vram so i added more vram to your vram:

Chonk has now picked up and put his very first kids toy away, fully autonomously! Many more to come!

only reason this guy isn't a millionaire is he spent all his money on RGB lighting

Lets see what this GB300 can do! This is Asus' variant of the DGX Station, they call it ET900N G3. More info and tests to come!

chonk just letting himself in like he owns the place already wtf

This cow side eyeing me this morn

btw this is a sales pitch and advertisement for the fact the GLM 5.3 family is frontier at home without some prick that thinks they know what's best for you hacking && cyber crimes are still very illegal and the penalties extremely severe it's just a good model stop sending this company and anyone like them money to advocate against you

Whats chonk got in his trunk

Thinking of callin this sourccey bot "chonk," what we think?

I lied. ONE MORE update. This is actually beginning to look acceptable and probably approaching maximal aesthetics given the limited waist DOF. wonder how it'd translate to real.

alright 1 more update on the asimov gait, think the reward function is just about where I want it. train it for a bit then work on the fall recovery.

some improvement on the gait. mjlab is just too fast to iterate with

porting over @asimovinc locomotion training to mjlab, looks like it's working. Now we need to fix this old man walk lmao

Acquiring honey

Bessent? The soybean farmer?

we're post economic anyways

After playing with the Jev hybrid model, I started looking into OpenJev, and started with a halite 1 implementation with: GLM 5.3 Flash + Jev [vs] GLM 5.3 Flash + OpenJev OpenJev is just a 4B model handily defeating Jev (in this tiny silly toy example ofc). Now I'm super curious to learn what Jev's actual model size is.

GLM 5.3 Flash vs Hybrid (GLM 5.3 Flash + Jev) on a turn-based strategy game (Halite 2) The hybrid approach is 13x faster, 56% of the pure GLM 5.3 Flash API cost, and gains a slight edge in performance. In the hybrid approach: -GLM 5.3 Flash handles strategic judgment of which planets matter, who expands, who fights, when to re-plan. -Jev handles the ship-by-ship execution. GLM 5.3 Flash alone defeats Jev alone 82% of the time. This is obviously just a single example and some of the first things I've tried to make use of Jev as an augmentation to an LLM. Could potentially make even more gains in cost, speed, and performance, but using what I've figured out here, I am compelled to try some ideas with robotics next.

it is shocking to me that there are still people trying to gatekeep programming by programming language. stop listening to people who speak this way, and just start building stuff.

you do have to at least respect the uptime for this new popular model

Is Irregular actually just an Israeli EA marketing firm? "Hire us, and we'll get your model to hack some companies" Which of the breakouts *didn't* involve Irregular? Which companies, if any, have cut ties due to all these breakouts that come from Irregular?

Fall food plot preppin

Ok 1 more pups pic. We have a couple anatolians too. Such a cool breed. Great for kids too when theyre outside Can also confirm they keep the coyotes, birds of prey, cats, snakes, and even rats away. They do a great job. Need to train for chickens tho they dont necessarily bond

You gotta watch out for these

We are so back! The RMA replacement GPU is here! Special thanks to @Coolmark482 for helping to push this thru!

A couple halite games with LLMs being given rules, states, and ability to choose moves. Here's DSV4.1 Flash vs GLM 5.3 Flash, both NVFP4 Going thru historical classical AI/logic games and using them to benchmark LLMs. Lots of interesting results. More to come!

vibe coding be like

I sure would love it if sharing live training runs would become the norm! It's just so cool to see.

it's probably nothing

GLM 5.3 Flash playin some starcraft 2

i did the native precision, documented here: hkinsley.com/reflections/all… and here: piped.video/PJBZmCy1Hfg Interestingly tho, I actually benchmarked the nvfp4 too and it got the *highest* score of any model lol. It's the new #1 on term bench 2.1 from my local tests w/ omp :o