Hugging Face AI Updates: May 8, 2026

1. ServiceNow-AI documents migrating an RL pipeline from vLLM V0 to V1

Hugging Face. ServiceNow-AI walks through migrating their RL training pipeline from vLLM V0 to V1, focusing on validating correctness before optimizing performance. The post covers inference-engine changes that affect reinforcement learning workflows in detail. Source

2. Open ASR Leaderboard adds private validation data

Hugging Face. The Open ASR Leaderboard now mixes private validation data into evaluation to combat benchmark gaming. The mechanism aims to keep speech-recognition rankings honest by preventing models from being tuned directly on public eval sets. Source