Topic: normalization

The ggml/llama.cpp project updated its models to use `rsqrt` normalization for GDN instead of `max`, and implemented an L2 norm for gated delta net q/k, enhancing model performance according to changes from flash-linear-attention. These updates are significant as they improve the efficiency and accuracy of the models used in natural language processing tasks.

A new dual-form ASR approach called Semantics-Aware Inverse Text Normalization has been developed for Chinese speech recognition, aiming to produce both accurate spoken-form transcripts and readable written-form transcripts. This method addresses the need in modern ASR scenarios where both forms are crucial for faithful transcription and readability.

arxiv.org↗