ByteDance Seed fixes full-pipeline FP8 RL instability with Calibrated Clipping across 8B-32B models
ByteDance-Seed · hf · 2026-09-23
ByteDance Seed's new research tackles a previously overlooked failure mode in full-pipeline FP8 reinforcement learning for LLMs: compounded quantization noise distorts the importance ratio, pushing negative-advantage tokens outside the trust region and erroneously zeroing their gradients, so pathological outputs accumulate as entropy surges and garbled text.
- Prior fixes (e.g. TIS) addressed train-inference mismatch, but instability persisted across the full FP8 pipeline.
- The proposed Calibrated Clipping aligns FP8 clipping bounds with high-precision BF16 distributions by matching the lower-bound clipping quantile and rebalancing the upper bound.
- Experiments span GRPO and DAPO, model scales from 8B to 32B, and multiple FP8 scaling granularities, eliminating entropy spikes and matching BF16 baseline performance.
More from Infra
- Gas-for-AI-datacenters forum co-located with CoRL shows energy demand spillover — carlosdponx · 2026-11-12
- a16z podcast: Mike Volpi on Cisco's 70+ acquisition playbook and the AI buildout — a16z · 2026-11-12
- AMD to pass Intel in server revenue by mid-2027 as share gains 6 points a year — BenBajarin · 2026-11-12
- Study: Reflective Data Center Roofs Actually Cool Daytime Ground by 1.8-2.2°C — SumitGup · 2026-11-12
- Tsinghua's BaGuaLu-2 tames MoE expert-load spikes at runtime, accepted to SC26 — jiqizhixin · 2026-11-12
- antirez backs low-bit DeepSeek 4.1 Flash over Qwen 27B dense on Apple M5 Macs — antirez · 2026-11-12