← Thursday, October 15, 2026InferencePaper

KV4-RL protects selected KV positions to speed INT4 rollouts

huggingface.co by Sihwan Park et al. Editor score 8/10

KV4-RL keeps attention sinks and recent context in INT8 while quantizing the rest to INT4, achieving 1.80× cumulative rollout speedup and 1.27× full-iteration speedup with final accuracy within 0.5 percentage points of BF16. The results show how selective KV-cache protection can improve RL training throughput while limiting quality loss.

What

- KV4-RL combines calibration-free INT4 KV-cache quantization with selective INT8 protection for attention sinks and recent context.

Method

- Protecting a bounded number of positions reduces sampler mismatch while retaining most INT4 speedup; INT4 alone can degrade learning even with TIS.

Why it matters

- In GRPO training, KV4-RL achieved 1.80× cumulative rollout and 1.27× full-iteration speedups over BF16, with final accuracy within 0.5 points on two models.

Read on huggingface.co →

Headline, summary, brief and diagram are written by a model from the source's own text; the link, byline and numbers come from the source. Read the original before relying on a claim.