LodeHQ
AI news for people who shipFree, by email

AI · developers.googleblog.com · Wednesday, 18 November 2026

Zero-TIM makes TPU RL truly on-policy at the bit level

Google’s Zero-TIM implementation removes the numerical gap that can make TPU-based LLM reinforcement learning quietly off-policy, giving training teams bitwise-identical logprobs from the sampler and trainer at the same weights. It aligns the arithmetic across decoding, serving prefill, and trainer forward passes, where different tensor shapes and floating-point reduction orders previously changed rounding. Tunix verified the approach across 64 TPU v5p chips in a DP8×TP8 setup.

Read the source: developers.googleblog.com ↗In the 18 November issue →