Paper Claims bf16 Breaks VLA Fine-tuning, Sparking Debate
A Google DeepMind researcher highlighted a paper showing bf16 precision breaks VLA fine-tuning, calling it one of the month's most interesting robotics papers. Community members pushed back, noting the issue is well known and can be mitigated with fp32 master weights or stochastic rounding.
2026-11-18 ~ 2026-11-18 · 2 related posts
- bf16 breaks VLA fine-tuning: one of the month's most interesting robotics papers — m_wulfmeier · 2026-11-18
- bf16 breaking VLA fine-tuning sparks debate: fp32 or stochastic rounding already known? — giffmana · 2026-11-18