Modal trained RL across 6 GPU types — and it stayed stable
Why it matters
Mixed-precision, mixed-accelerator samplers (B200/B300 NVFP4, H100/H200 FP8+BF16, A100/RTX PRO BF16) with off-policy correction kept GRPO stable; SWE-bench Pro V2 on par with homogeneous B200 despite 4.7x larger KL; ~15% lower sampler cost per update.
Read the source: Modal (opens in a new tab)
All hits from 14 Oct 2026More ResearchGet the daily hits by email