Crimson AI NewsA CrimsonLingua Network service
EN ع
← Back to news
Research paper Hugging Face

Hugging Face Researchers Tackle RL Instability in Small Language Models

AI By Crimson AI Hugging Face Papers 29 July 2026 · 00:00 13 views
Share: X Telegram

A new study from Hugging Face identifies three reproducible failure modes in reinforcement learning for small language models (70M–500M) and proposes a robust training pipeline that stabilizes PPO, improving win rates over supervised baselines with less data.

Hugging Face Researchers Tackle RL Instability in Small Language Models

Key points

Reinforcement learning (RL) is a powerful tool for aligning language models, but applying it to small language models (SLMs) in the 70–500M parameter range has often proven unstable. A new paper from Hugging Face, presented at the 2026 IEEE International Conference on Systems, Man, and Cybernetics (SMC), systematically investigates these failures and offers a robust solution.

The researchers trained fifteen (model, corpus) configurations using Proximal Policy Optimization (PPO), spanning models like Pythia-70M/160M/410M and SmolLM2-135M/360M on datasets such as TinyStories, CNN/DailyMail, and Wikitext-103. They identified three reproducible failure modes: silent LoRA parameter freezing in standard PEFT/TRL pipelines, numerical overflow in importance ratios when using bfloat16, and catastrophic policy collapse due to reward-model error.

To counter these issues, the team introduced a merge-and-reinitialize adapter technique, switched to float32 precision during PPO updates, and implemented a three-layer safety mechanism comprising reward whitening, importance-ratio guarding, and weight rollback. This pipeline led to stable convergence across all experiments.

The paper also proposes a "capacity-headroom hypothesis": PPO performance at the SLM scale depends on both a fluent supervised model (perplexity

All checkpoints, preference datasets, and training scripts have been publicly released to support reproducibility and further research.

ModelCorpusFailure ModeMitigation
Pythia-70M/160M/410MTinyStories, CNN/DailyMail, Wikitext-103Silent LoRA freezing, bfloat16 overflow, policy collapseMerge-and-reinitialize adapters, float32, safety layers
SmolLM2-135M/360MTinyStories, CNN/DailyMail, Wikitext-103Silent LoRA freezing, bfloat16 overflow, policy collapseMerge-and-reinitialize adapters, float32, safety layers
Source
Hugging Face · Hugging Face Papers
Related news
Research paper
Hugging Face 31 Aug 2026

Hugging Face Unveils StepGuard: Step-Level Guardrails for Safer AI Agents

StepGuard, a new step-level guard model from Hugging Face, audits agent actions before execution, reducing attack success rates by...

1
Research paper
Hugging Face 31 Aug 2026

Hugging Face Researchers Unveil ABot-Recon for Stable Long-Horizon 3D Reconstruction

ABot-Recon, a new streaming 3D reconstruction model from Hugging Face, achieves stable long-horizon performance using only local t...

1
Research paper
Hugging Face 31 Aug 2026

ContextPilot: Teaching Agents Proactive Context Management via Fine-Grained RL

Hugging Face researchers introduce ContextPilot, a framework that enhances long-horizon agent reasoning by expanding context-editi...

1