SearcharxivSearch

EXPLORE CONNECTIONS

Decomposing MXFP4 quantization error for LLM reinforcement learning: reducible bias, recoverable deadzone, and an irreducible floor

Follow the relationships that help you find your next source.

Connections are not prepared for this record yet, or its source metadata has changed. The original record remains available while background snapshots are rebuilt.