Qeravio
Canonical AI event

GGML replaces GGML_FA_ALL_QUANTS with GGML_FA_QUANTS for more control.

CUDA now supports configurable quant combinations and runtime fallback warnings for uncompiled combinations.

9 Sept 20261 verified claims1 sources1 observations
What happened

The official source reports this update: b10876: CUDA: replace GGML_FA_ALL_QUANTS with GGML_FA_QUANTS, more control over what is compiled.

Why it matters

This official update documents a development concerning b10876: CUDA: replace GGML_FA_ALL_QUANTS with GGML_FA_QUANTS, more control over what is compiled. 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
infrastructureLlama project changes return type to int32_t for sampler_chain_n.9 Sept 2026infrastructureLlama.cpp officially deprecates --mmap, --mlock, and --dio arguments.9 Sept 2026infrastructureIBM fixes granite3 moe unknown parameter count in model.9 Sept 2026infrastructureLlama.cpp adds video ID propagation to bitmap feature9 Sept 2026infrastructureJinja updates null left operand handling in membership tests.9 Sept 2026infrastructureVulkan adds dedicated iq4_xs mat-vec shader for RDNA4, improving token generation.9 Sept 2026
Evidence trail

Sources behind the event

llama.cpp official releasesb10876: CUDA: replace GGML_FA_ALL_QUANTS with GGML_FA_QUANTS, more control over what is compiled
Editorial presentation

Read the public story

Open story