Oriel — AI & technology worth noticing, every day.

InfraRSI: Harnessing Recursive Self-Improvement for Training Infrastructure

Paper · https://arxiv.org/abs/2611.11558

While Recursive Self-Improvement (RSI) accelerates isolated algorithm and kernel synthesis, distributed LLM training depends on tightly coupled launch settings, operator kernels, and communication schedules, where single edits often hit out-of-memory (OOM) boundaries, compiler conflicts, measurement jitter, or silent numerical drift. We present InfraRSI, the agentic harness that turns open-ended human commands into verified training optimizations via a five-stage workflow: init locks editable scope and a fixed task contract over workload, loss, and timing invariants to prevent reward hacking, doctor profiles baseline phases and calibrates cluster noise, run co-optimizes launch knobs and code patches before re-verifying top candidates against the baseline and logging findings to a cross-hardware experience bank, and report and apply audit source paths, track failed-trial costs, and commit promoted changes. Rather than stepping greedily from a single incumbent, InfraRSI initializes from diverse experience-seeded configurations and fits global execution-time and censored peak-memory models across all trials, including OOMs. Using these models, it prices per-setting memory costs, revises miscalibrated LLM priors from live runs, and pairs memory-hungry optimizations with memory savers and verified code patches. Across four single- and multi-node dense and Mixture-of-Experts (MoE) LLM workloads, InfraRSI achieves a speedup of up to 2.75times and lifts mean Model FLOPs Utilization (MFU) from 17.4% to 30.0%, outperforming existing tuners and coding agents while transferring experience across GPUs. As it improves training infrastructure efficiency, InfraRSI forms a two-level Recursive Self-Improvement loop---revising priors online within runs and distilling experience across clusters---to push foundation model scaling.

First seen by Oriel on .

AI summary

InfraRSI is an agentic harness that converts open-ended commands into verified distributed LLM training optimizations, achieving up to 2.75× speedup across dense and Mixture-of-Experts workloads.

  • Its five-stage workflow constrains editable scope, profiles baseline performance and cluster noise, co-optimizes settings and patches, audits results, and commits promoted changes.
  • Global execution-time and censored peak-memory models use every trial, including OOMs, to combine memory-intensive optimizations with memory savers and verified code patches.
  • Across four single- and multi-node workloads, mean Model FLOPs Utilization increased from 17.4% to 30.0%, outperforming existing tuners and coding agents.
  • InfraRSI revises LLM priors using live runs and stores verified findings in a cross-hardware experience bank that transfers knowledge across GPUs and clusters.

#llm training #recursive self-improvement #distributed systems #performance optimization

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Points

Score reported by the source.

  • Hugging Face Papers: 33, latest observation ; +16 over observed period since

Comments

Comment count reported by the source.

  • Hugging Face Papers: 6, latest observation ; +1 over observed period since

Aliases

SourceExternal IDURL
Hugging Face Papers2611.11558https://arxiv.org/abs/2611.11558 Read discussion