Baseten powers Mirendil's frontier RL with ~40s weight syncs and 1.9x throughput
baseten · x · 2026-11-03
Baseten details its collaboration with Mirendil, the self-improving AI lab founded by ex-Anthropic researchers Behnam Neyshabur and Harsh Mehta with a $200M seed round. Rollout fleets can use 2-5x the trainer's GPUs, so Baseten watches Mirendil's checkpoint stream, reconstructs new policies, syncs weight deltas in under 40 seconds across global clusters, and delivers 1.9x higher sampling throughput.
More from Infra
- Weight teleportation open-sourced: cross-datacenter RL weight updates under 0.23s — gharik · 2026-11-03
- YouTuber's "F-You to Sam Altman" video showcases a local AI GPU rig — unchikuso · 2026-11-03
- Is compute fungible enough for a commodity market? Skeptic says chips and networking don't move — toptickcrypto · 2026-11-03
- Bittensor Subnet Engy Undercuts CoreWeave Inference Pricing by Up to 89% — markjeffrey · 2026-11-03
- Prefrontal open-sources LSA: 72.4x faster attention prefill at 10M tokens, accuracy on par — jcrwhittington · 2026-11-03
- DeepSeek-V4.1-Flash Throughput Benchmarked Across 3 Tools on 4x RTX PRO 6000 — TheZachMueller · 2026-11-03