TL;DR: llama.cpp's b9987 release introduces new GGUF tensor shape accessors, simplifying direct access to tensor dimensions within GGUF files.
Summary: The latest llama.cpp release, b9987, adds new GGUF tensor shape accessors. This update includes gguf_get_tensor_ne for retrieving tensor shapes as const int64_t *, while removing the n_dims accessor. These changes streamline how developers can inspect and utilize tensor dimensions directly from GGUF files.
Why it matters: This improves interoperability and ease of development for tools working with GGUF models, especially for custom inference engines or model analysis. Builders should explore how these new accessors can simplify their GGUF parsing and manipulation workflows.
Source: github_releases