Skip to content

HiLoRe: What to Store, Compress, or Recompute for Efficient GRPO Training

Sep 2026 · 0 citations · 50 references
Computer Science

TL;DR

HiLoRe is introduced, which allocates graph-attributed recovery units among high-precision storage, low-precision compression, and deterministic recomputation using measured recovery utility and update-conditioned approximation risk, and combines high-precision storage and deterministic recomputation with low-precision recovery under a calibrated risk budget.

Abstract

Group-relative policy optimization (GRPO) makes learner-side activations a major memory-computation bottleneck: gradient checkpointing reduces activation memory through recomputation, but fixed schedules can leave roughly 18 GB unused on a 48-GB GPU despite substantial recomputation overhead. Existing activation-management methods set state fidelity from execution cost, tensor properties, or generic compression sensitivity, without explicitly incorporating GRPO's analytic update structure into state-fidelity allocation. We formalize this dependence as policy-update exposure, linking the current GRPO loss coefficients to state-level approximation sensitivity. These coefficients are available before backward without an additional backward pass. We introduce HiLoRe, which allocates graph-attributed recovery units among high-precision storage, low-precision compression, and deterministic recomputation using measured recovery utility and update-conditioned approximation risk. It combines high-precision storage and deterministic recomputation with low-precision recovery under a calibrated risk budget. Across five model-task settings with 2K responses and memory<1.10 times GC's per-GPU actor-update peak, HiLoRe's actor-update throughput gains reach 13.5% over GC and 7.9% over the fastest evaluated baseline, with paired mean downstream-score differences below 0.6 percentage points.

View source

Similar papers

Sep 2026

AReaL-TIK: Stateful Agentic Optimization of Unified RL Kernels through an Optimization IR

Reinforcement learning (RL) post-training often uses distinct GPU kernels for rollout and policy update. In synchronous PPO and GRPO, numerical disagreement can perturb ratios between current token probabilities and those assigned during rollout. Recomputing rollout log-probabilities with the policy-update backend avoi...

Ran Yan, You-He Jiang, Jia-Yi Nie et al. · 0 citations
#machine learning Preprint Oct 2026

Flash-OPD: Fast On-Policy Distillation

On-policy distillation (OPD) provides dense teacher supervision on student-generated trajectories, but generating and evaluating long rollouts incurs substantial training cost. Existing acceleration methods reduce this cost through open-loop rollout schedules or closed-loop horizon adaptation. However, supervision comp...

Wei Chen, Junle Chen, Yi-Tong Yang et al. · 0 citations
Preprint Aug 2026

LazyTrain: Limited-resource Allocation toward Zero-waste Yield Optimization in Large Language Model Training

LazyTrain is proposed, an optimization layer over a layer-streaming executor that formulates checkpoint selection, activation placement, recomputation, and CPU-GPU-NVMe communication overlap as a mixed-integer scheduling problem, then executes the solved policy during training.

Xiao-Jun Wu, Ce-Hao Yang, Hong-Hao Liu et al. · 1 citation
Preprint Aug 2026

Benchmarking Composable Compression Techniques in Mixture-of-Experts LLMs

The results reveal non-trivial interactions among compression methods: composable compression cannot be predicted from standalone techniques, compression rate alone does not reliably predict quality loss or runtime gain, expert pruning is the dominant degradation source, and average quality can hide workload and archit...

A. Benazir, Chen Chen, Rongxiao Qu et al. · 0 citations
Preprint Sep 2026

Explicit State and Resource Contracts for Low-Precision Pipeline Parallel Training under Captured Graphs

CUDA Graphs eliminate launch overheads by replaying tensor operations over static virtual addresses. However, FP8 pipeline training continuously alters the scaling states, microbatches, and deferred backward tasks that those fixed addresses represent. Split-backward schedules (e.g., 1F1B, Zero-Bubble) decouple input-gr...

Gen-Lang Chen, Jun-Yi Zhu · 1 citation
#large language models Book Open access Aug 2026

NeuroPrefetcher: Storage-Aware Sparse LLM Inference via Delta Prefetching

NeuroPrefetcher is presented, a storage-backed LLM inference system that exploits that MLP activity during autoregressive decoding has strong temporal locality, and achieves 7.9-12.0x speedup over llama.cpp across constrained memory budgets.

Nobel Dhar, Md Romyull Islam, Xue-Chen Zhang et al. · 0 citations

Related blog posts

Microsoft Research Blog Nov 16, 2026

TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient

AI coding agents are increasingly help developers fix bugs and build features largely on their own, but every interaction with their models consumes tokens. In each session, an agent sends instructions, tool descriptions, and session history to its model. Most efforts to reduce this cost focus on individual tasks. Yet developers return to the same […] The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.