LMSYS Debuts Miles: Blackwell-Native 8-bit and 4-bit RL Recipes
BanghuaZ · x · 2026-07-30
LMSYS has introduced two Blackwell-native, low-precision reinforcement learning (RL) recipes within the Miles framework, fully open-sourced in collaboration with @humansand and @nvidia.
Key technical highlights include:
- End-to-end MXFP8 across rollout, forward, and backward GEMMs.
- Hardware-native NVFP4 W4A4 RL specifically optimized for MoE models.
- Bit-exact quantizers between training and rollout to minimize mismatch.
- Fine-grained, cross-stack per-layer precision control.
Testing on Qwen3-30B-A3B shows that all 5 low-precision configurations closely track the BF16 reward curve while significantly reducing rollout time.
More from Infra
- Reflection AI open-sources Beam under Apache 2.0 with 40+ ecosystem partners — zhyncs42 · 2026-10-16
- NVIDIA launches RTX Spark for local AI on Windows PCs — ostrisai · 2026-10-16
- SkyRL ships Day-0 RL support for Beam with +62% LoRA decode throughput in vLLM — vllm_project · 2026-10-16
- DFlash2 delivers up to 3.03x decode-only speedup for Beam across five benchmarks — vllm_project · 2026-10-16
- vLLM's Beam serving kernels cut GB200 forward time from 8.15 to 6.47 ms at TP8 batch 128 — vllm_project · 2026-10-16
- Reflection's 501B MoE Beam hits 463 tok/s on 8x GB200 with NVFP4 at day 0 — zhyncs42 · 2026-10-16