Gabor Csapo retweeted
More compute doesn't always mean it's better.
Catch @leavittron, co-founder of @datologyai, at Backprop Summit to hear why data quality could be the real multiplier behind better models.
luma.com/backprop
Gabor Csapo retweeted
When a model rephrases documents the knowledge cutoff guarantees are lost. It can introduce information its sources never had. This is catastrophic for backtesting. So we built a way to generate synthetic pretraining data with measurable, controllable knowledge cutoff guarantees
Gabor Csapo retweeted
AI filmmaking is trash.
I mean, AI filmmaking is real filmmaking now.
Today we're launching Martini Bar from @martini_film: production infrastructure for AI filmmaking companies. 🍸
Gabor Csapo retweeted
Applications are open for the January cohort of the Datology Labs Fellows Program.
Funding, compute, and mentorship to work on the science of data.
Paid · One semester · San Mateo/NYC
Apply by Nov 15: labs.datologyai.com/fellowsh…
Zephon’s TorchTitan integration preserves identical global batches when training resumes on a different GPU count.
In our latest blog’s 1B-parameter model test, training resumed from a 16-GPU checkpoint on 8, 32, and 64 GPUs, and the SHA-256 hash of every global batch matched the original run.
Zephon tokenizes, packs, shuffles, and mixes data inside the training job, avoiding a preprocessed copy of the corpus.
The data, tokenizer, seed, mixture, context length, and global batch size must stay unchanged when resuming. The lane count must divide evenly across data-parallel replicas, and each optimizer step’s batch must split evenly across lanes. Training losses can still differ because floating-point reduction order depends on GPU count.
✍️ @datologyai @josh_wills @MaxiBoether, Ties Robroek (@datologyai)
Read more: pytorch.org/blog/bringing-el…
For all the data researchers out there!
Today we’re launching Datology Labs: a new commitment to open and collaborative research into the science of data.
Research has been a core pillar of DatologyAI from the start. Now we’re partnering with researchers at universities and across industry to take it further.
Launching Datology Labs
12
269
Gabor Csapo retweeted
Today we’re launching Datology Labs: a new commitment to open and collaborative research into the science of data.
Research has been a core pillar of DatologyAI from the start. Now we’re partnering with researchers at universities and across industry to take it further.
Launching Datology Labs
Gabor Csapo retweeted
Our curation workflows (DAGs) have 50+ configurable stages & operators. A large run expands into 540+ unique operator configurations and 2460 actual distributed processing jobs.
Many operators had to be built from scratch because vanilla Spark operations simply weren't practical at this scale.
Gabor Csapo retweeted
Nothing. Actually nothing. moves the needle on model performance as much as improvements in data quality. You probably can’t trust your vendors, your data is probably a mess. You need tools to curate data yourself. @datologyai is making this possible.
Gabor Csapo retweeted
Opinions expressed in these memes are my own and not necessarily those of my employer.
But if I had to guess, they're probably also the views of my employer. Curate your data, y'all.
Introducing Zomma Agent Fleets 🙂↕👇
It's 2026. Why is your customer ticket still stuck for days?
Because most teams still work in a single queue.
With Zomma, you can now dispatch a Fleet of parallel computer-use agents as soon as work comes in.
The biggest hack? Fleet = better outcome with fewer tokens. Each subagent gets just enough context to do the work well, saving tokens and time.
And you can score every subagent against your own eval set.
@ZommaLabs is in early access. If you're building agents for your operations, let's chat 📥
Gabor Csapo retweeted
New from DatologyAI: Zephon, an open-source data loader for text and multimodal training.
Resume a run on a different number of GPUs and most loaders quietly change the data. In our tests, GPU count alone moved eval scores by up to 0.82 points, about 4x the gap used to choose between data curation methods.
tl;dr: same training batches on 8, 16, 32 or 64 GPUs, even while tokenizing, packing and mixing on the fly.
Gabor Csapo retweeted
Look at him, that’s my 1000x Engineer
Your what?
My 1000x Engineer! My throughput whisperer. Look at him, you notice anything different about him? Look at his face
I’ll give you a hint, his name is Josh. He put H100 in his name before it was cool. Yeah, I’m sure Zephon scales.
Data Loader (n.): The part of your training stack that is too ML-y for the systems folks and too systems-y for the ML folks.
It's the work I love best, and I had a ton of fun building a new one with the team here at @DatologyAI. My three favorite things about Zephon:
Gabor Csapo retweeted
At @datologyai we run thousands of data experiments. Two things keep biting us:
- Same run on 16 vs 8 GPUs → different evals
- Every pipeline change → re-materialize the dataset
Introducing Zephon, the data loader we built to process data on the fly, deterministically. 🧵
Gabor Csapo retweeted
Picture stepping into a live 3D broadcast or 3D video call and looking around freely. That is the promise of dynamic NVS.
But multi-view camera setups are fixed so parts stay hidden. Looking anywhere means recovering them from memory, which current stateful and stateless models cannot do in real-time over long-horizons.
Until now. Introducing NSTM.
⚡️Real-time inference + 🧠minute-long memory.






















