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

Reasoning in Greek: SFT Shifts Language, RL Fixes Defects, Accuracy Misses Both

AI By Crimson AI Hugging Face Papers 21 August 2026 · 00:00 9 views
Share: X Telegram

A new study on fine-tuning MoE models to reason in Greek finds accuracy is a noisy, blind metric; SFT builds fluency while RLVR repairs formatting and leakage defects.

Reasoning in Greek: SFT Shifts Language, RL Fixes Defects, Accuracy Misses Both

Key points

A new research paper from Hugging Face investigates what happens when large mixture-of-experts (MoE) models are fine-tuned to reason in a low-resource language—Greek. The headline finding is a null result: accuracy barely changes, and the benchmark itself is too noisy to detect meaningful differences at this scale. Simply changing the random seed swings scores by 7.7 points, more than any data or recipe effect measured.

The real transformations occur where accuracy cannot see. Base models never reason in Greek—0 out of 1,000 traces—even when the question is in Greek. After supervised fine-tuning (SFT), all released checkpoints reason in the question's language on ~98% of items, with one family using 3× fewer tokens. Grammaticality improved on all four models, and general ability stayed within a few points of the base, indicating no catastrophic forgetting.

However, SFT cannot fix its own defects: a quarter of answers skip the requested format, answers leak into the reasoning channel, and an explicit 'think in English' instruction is obeyed less than half the time. Reinforcement learning with verifiable rewards (RLVR), pre-registered before training, fixes the first two outright—format fallback drops from 24% to 2.5%, and leakage from 3.5% to 0.0%—and improves instruction following by 9.1 percentage points, all against a flat random-reward control.

The authors propose six behavioural dimensions to make such changes measurable, each gated to reject metrics that correlate with output length, and they document six failures of their own instruments, each caught by a control. They release five checkpoints, and stress that the instruments, controls, and pre-registration transfer to any low-resource language.

MetricBefore SFTAfter SFTAfter RLVR
Reasoning in Greek (traces)0/1,000~98%~98% (survives)
Format fallback24%2.5%
Answer leakage3.5%0.0%
Obey 'think in English'+9.1pp
Source
Hugging Face · Hugging Face Papers
Related news
Research paper
Hugging Face 29 Aug 2026

Hugging Face Audit: 110 of 124 AI Evaluations Fail to Support Their Claims

A new commit-bound census of 124 Inspect Evals units reveals that 110 stop before deterministic inference due to missing historica...

4
Research paper
Hugging Face 29 Aug 2026

Aphanta: New Framework Diagnoses When Image Editing Boosts Multimodal Reasoning

Hugging Face researchers introduce Aphanta, a diagnostic framework that evaluates when image-editing intermediates improve multimo...

5
Research paper
Hugging Face 29 Aug 2026

EditaLive! Enables Real-Time Character Video Editing for Live Streaming

Hugging Face researchers introduce EditaLive, a framework for real-time human-centric video editing in live streams, achieving sta...

4