Hugging Face AI Updates: May 7, 2026

1. ServiceNow Engineers Walk Through vLLM V0 to V1 RL Migration

ServiceNow. A Hugging Face engineering post by ServiceNow AI documents migrating reinforcement learning workloads from vLLM V0 to V1, prioritizing correctness over premature performance fixes. The piece details determinism issues, sampling drift, and the test suite they used to confirm parity before flipping production. Source

2. Open ASR Leaderboard Adds Private Eval Set to Curb Benchmark Gaming

Hugging Face. The Open ASR Leaderboard is adding a private “benchmaxxer repellant” evaluation set to fight overfitting on public test data. Public scores will continue to be reported, but a hidden split now backs the published rankings. Source