Skip to content
Tagged articles

RL post-training

4 articles · Page 1 of 1
Weekly Large Model Application
Weekly Large Model Application
Oct 12, 2026 · Artificial Intelligence

Designing a Telephone-Specific Chinese ASR: Base Model + RL Post-Training Wins

This article presents a system design for a Chinese telephone-specific ASR model, explaining why generic ASR fails on calls due to 8kHz narrowband, codec distortions, and overlapping speech, and proposes a pluggable base model with a four-source data engine, RL post-training using verifiable rewards like entity F1, role-aware decoding, and layered evaluation focused on business-critical metrics rather than raw CER.

ASRChinese ASRRL post-training
0 likes · 12 min read
Designing a Telephone-Specific Chinese ASR: Base Model + RL Post-Training Wins
Machine Heart
Machine Heart
Jul 12, 2026 · Artificial Intelligence

Does One Update Really Strengthen a Policy? PIRL and PIPO for Closed‑Loop RL

The paper by researchers from Beihang, Peking University and Meituan proposes PIRL, a new RL‑post‑training perspective that treats policy improvement as the optimization objective, and PIPO, a plug‑and‑play framework that adds a verification loop to amplify beneficial updates and suppress harmful ones, demonstrating consistent gains across math reasoning, code and tool‑use tasks.

Importance SamplingPIPOPIRL
0 likes · 9 min read
Does One Update Really Strengthen a Policy? PIRL and PIPO for Closed‑Loop RL
Machine Heart
Machine Heart
Jun 21, 2026 · Artificial Intelligence

Why Post‑Training Makes Large Reasoning Models Overconfident and How LED Restores Exploration

The paper reveals that reinforcement‑learning post‑training flattens the entropy of the final layer in large reasoning models, making higher sampling temperatures ineffective, and introduces Latent Exploration Decoding (LED) to recover exploration from intermediate layers, yielding consistent pass@k gains without extra training.

LED methodRL post-trainingentropy collapse
0 likes · 13 min read
Why Post‑Training Makes Large Reasoning Models Overconfident and How LED Restores Exploration
Machine Heart
Machine Heart
May 13, 2026 · Artificial Intelligence

Why Bigger Teachers Don’t Teach Better: Tsinghua’s On‑Policy Distillation Study

Recent research by Tsinghua and collaborators dissects On‑Policy Distillation for large language models, revealing that higher‑scoring teachers often fail to improve students unless their thinking patterns align, detailing token‑level overlap dynamics, failure cases, and two practical remedies to rescue ineffective distillation.

Model ScalingOn-Policy DistillationRL post-training
0 likes · 9 min read
Why Bigger Teachers Don’t Teach Better: Tsinghua’s On‑Policy Distillation Study