Introducing RL Swarm 72B

Fully decentralised RL training of 72B-parameter models for anyone to join, with no whitelists.

Train your base model on a new advanced math dataset (DAPO-Math-17k) collaboratively alongside thousands of others using our novel multi-stage system.

The network for machine intelligence


https://gensyn.ai

Update your node now and choose a swarm:

Consumer (>8GB VRAM): now even lower-powered devices can run the swarm (GSM8K)

More powerful: bring a bigger model, up to 72B params, to tackle a harder dataset (DAPO-Math-17k)

Also in this release, we’ve added:

- multi-peer ID / EOA mapping - allowing you to link multiple nodes to a single EOA with the same email address.

- better reward and participation tracking, with new dashboard versions

- memory optimisations for consumer devices

Update your node now and choose a swarm:

Consumer (>8GB VRAM): now even lower-powered devices can run the swarm (GSM8K)

More powerful: bring a bigger model, up to 72B params, to tackle a harder dataset (DAPO-Math-17k)

https://github.com/gensyn-ai/rl-swarm

We’ve long held that artificial superintelligence will come from a collection of diverse models that interoperate, rather than a single monolith.

Multi-swarm on the Gensyn Testnet is the first step in this direction, with the new swarm running alongside the original.

Does it really matter if we run a node for RL Swarm Math or Hard Math?