OossaAI is evolving fast. We explain it simply.

llama.cpp adds classifier pooling support for rerankers

The llama.cpp library now handles classifier_pooling, improving ModernBERT reranker use on many platforms.

NoteOossa1 min read

The open‑source llama.cpp project released a new version that adds support for classifier_pooling, a setting used by ModernBERT rerankers. The change lets the library read and write the pooling type, defaulting to mean when unspecified. It also renames the internal field to pooling_type_cls. The update is available for a wide range of OSes and hardware, from Apple Silicon Macs to Linux GPUs and Android devices.

Why it matters

It makes llama.cpp more usable for search and ranking tasks across many devices.

Was this article useful?
Share

Read next

Oossa · Newsletter

The week in AI, explained

Every Monday: the stories worth knowing, in plain language. Free, no spam.

Sources & references

#SourceOutletDateKey takeaway
1ggml-org/llama.cpp b11273 ↗llama.cppSep 30, 2026<details open> model : support classifier_pooling for rerankers (#29627) * model : support classifier_pooling for ModernBERT rerankers Assis

1 sources

Last updated: ·Markdown·llms.txt