Qeravio
Canonical AI event

Fused QKV gemma4, qwen35 models now calculate split states for attention.

Fixes for fused QKV gemma4 and qwen35 models now support macOS and iOS.

17 Sept 20261 verified claims1 sources1 observations
What happened

The official source reports this update: b11009: TP: fix split state and granularity for fused QKV gemma4, qwen35. TP: fix split state and granularity for fused QKV gemma4, qwen35 ( #28965 ) model: calculate split states for attn_qkv from n_head * n_embd_head_k required for gemma4 with --fuse-qkv, where n_embd is 5376 but Q is 8192.

Why it matters

This official update documents a development concerning b11009: TP: fix split state and granularity for fused QKV gemma4, qwen35. 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
infrastructureHexagon adds support for K-Quants Q4_K and Q6_K kernels17 Sept 2026infrastructureCUDA graph usage improved for MTP in llama.cpp draft17 Sept 2026infrastructureLlama app now supports zeroed rope probe in decoding graph.16 Sept 2026infrastructureLlama.cpp improves im2col access patterns in CUDA/HIP16 Sept 2026infrastructurespacemit fixes int16 data transpose bug in permute_transpose_impl16 Sept 2026infrastructureServer discards cached graphs on buffer free to prevent remote code execution.16 Sept 2026
Evidence trail

Sources behind the event

llama.cpp official releasesb11009: TP: fix split state and granularity for fused QKV gemma4, qwen35
Editorial presentation

Read the public story

Open story