Qeravio
Canonical AI event

Llama server reduces image processing time by up to 7.61x with scheduler optimization.

The update reduces processing time for multi-image and video inputs on Gemma models by eliminating unnecessary scheduler re-reserves.

29 Sept 20261 verified claims1 sources1 observations
What happened

The official source reports this update: b11227: context : do not re-reserve the scheduler when toggling causal_attn. context : do not re-reserve the scheduler when toggling causal_attn ( #28751 ) context : do not re-reserve the scheduler when toggling causal_attn llama_context::set_causal_attn() marks the scheduler to do a full re-reserve on every change of the flag. For vision inputs, this flag is flipped twice around each non-causal image chunk for Gemma models, resulting in two expensive sched_reserve() passes per image. This is especially slow for multi-image or video inputs.

Why it matters

This official update documents a development concerning b11227: context : do not re-reserve the scheduler when toggling causal_attn. Its practical significance depends on the scope and evidence stated by the source.

What to watch next

Read the official source update and verify its stated scope, evidence and timing before acting on it.

Connected knowledge

Entities affected by this event

Llama Cppsubject
Continue this topic
infrastructureGgml-cpu adds AVX2 support for masked loading and storing on x86.29 Sept 2026infrastructureVulkan compile error fixed in llama.cpp update29 Sept 2026infrastructureGemma 4 audio embeddings now use ggml_pad_ext for left padding.29 Sept 2026infrastructureggml-openvino disables unaligned batch-stride views29 Sept 2026infrastructureHugging Face fixes HF cache paths on Windows29 Sept 2026infrastructureLlama.cpp now handles unaligned writes in WebGPU buffer.29 Sept 2026
Evidence trail

Sources behind the event

llama.cpp official releasesb11227: context : do not re-reserve the scheduler when toggling causal_attn
Editorial presentation

Read the public story

Open story