Pinned post
Not all data centers are born equal.
Liquid training allows complex AI workloads to run across compute around the world.
Liquid training allows complex AI workloads to run across compute around the world.
0:37
crux
Verified account This account is verified. Learn more@macrocrux
CTO & Co-Founder @ Macrocosmos
253 Following 2.7K Followers 690 Posts
Latest posts 5
Can’t believe it’s almost 2027 and small changes to the prompt actually matter any more. Are we not past prompt engineering?
If this is true in general it explains why I always feel like claude understands what i want better when im using voice mode — I guess the transcript is (almost) always grammatically correct.
If this is true in general it explains why I always feel like claude understands what i want better when im using voice mode — I guess the transcript is (almost) always grammatically correct.
Nice DeepSWE score! Does it survive a lowercase filename?
Introducing MaxxBench: 89 DeepSWE tasks with small changes to the instructions or starting code. Same engineering problem, slightly different requirement.
Models clustered around 74% on DeepSWE end up 23 pp apart.
Good at coding, or good at DeepSWE? 🧵View quoted post on X
The use case i’m most excited about atm is agentic orchestration for running large GPU fleets. You have to schedule compute at scale, consume realtime logs & telemetry at the same time. Both ML- and cost-optimal performance depends on fast decision making.
Our current policy layer places machines on reliability, cost and location and it’s pretty rigid. We’re exploring an agentic layer for iota that assigns nodes within and across runs, diagnoses faults and works out what caused them.
Our current policy layer places machines on reliability, cost and location and it’s pretty rigid. We’re exploring an agentic layer for iota that assigns nodes within and across runs, diagnoses faults and works out what caused them.
first 24h of iota-decide-4b:View quoted post on X
@huggingface built a free demo space for it
@pratikg benchmarked it on a mac mini at ~90ms a decision
We’re testing it internally as an agentic control plane for @IOTA_SN9, so the swarm can diagnose and recover from its own failures in real time.
Incredibly proud that we have produced a pareto optimal model. It's lightning fast and very good.
IOTA Decide 4B began as an internal fine tuning experiment on our SDK, led by @v_truba.
We didn't need to wait for a reserved cluster, we just provisioned 16 GPUs from around the world and ran our workloads across them. Two years of nonstop grinding are paying off now.
Lot's more powerful SDK use cases to share very soon.
IOTA Decide 4B began as an internal fine tuning experiment on our SDK, led by @v_truba.
We didn't need to wait for a reserved cluster, we just provisioned 16 GPUs from around the world and ran our workloads across them. Two years of nonstop grinding are paying off now.
Lot's more powerful SDK use cases to share very soon.
Introducing IOTA Decide 4B, our first public model.
Pareto optimal. Built different.
Trained across globally distributed GPUs with our SDK.
Open weights, Apache-2.0.
🤗 huggingface.co/macrocosm-os/i…View quoted post on X
We'll know our work in distributed training is done when people stop caring that it's distributed.
Nobody should care whether a model was trained distributed. They should care that it’s useful.View quoted post on X
We’ve been using models we trained across distributed GPUs, and they’re very good. Sharing one soon.
