← ALL RECIPES
GOOGLE GEMMA 4PREVIEW

GEMMA 4 26B-A4B (MOE)

The 8-GB-card class leader: 26B-class quality with ~4B active parameters per token.

PARAMETERS

26B (3.8B active)

ARCHITECTURE

MoE

QUANT

QAT Q4_K_XL (13.2 GB)

VRAM

8 GB+ with expert offload

01

OVERVIEW

Gemma 4 26B-A4B routes each token through a small subset of experts (~3.8B active), and Google's QAT quantization keeps the 4-bit file at 13.2 GB with minimal quality loss — the community-standard pick for 8 GB GPUs.

IronWorks supports it with a hybrid path (CPU prefill + GPU decode) and a learned MoE expert cache: expert routing counts persist to a sidecar file, so the hottest experts are pre-warmed into VRAM on load (measured +40% over a cold cache on our hybrid-MoE models, 98–99% hit rates).

02

PREREQUISITES

  • 8 GB+ NVIDIA GPU with expert offload; 16 GB system RAM
  • 13.2 GB model file (QAT UD-Q4_K_XL)
03

DEPLOY

Install IronWorks

terminal
# CLI (CUDA build)
cargo install ironworks-cli --features cuda

# or build from source
git clone https://github.com/nervosys/IronWorks && cd IronWorks
cargo build --release -p ironworks-cli --features cuda

Run with a VRAM budget + expert cache

terminal
IRONWORKS_MOE_CACHE_GB=6 \
iwx complete -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --gpu \
  --vram-budget 7G \
  -p "Plan a two-week Japan itinerary." -n 512
04

TUNING

Learned expert pinning

Routing statistics persist to <model>.iwx_usage automatically when the expert cache is enabled; subsequent loads pre-warm the hottest experts.

terminal
# First run learns the routing distribution; later runs start warm.
IRONWORKS_MOE_CACHE_GB=6 iwx chat -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --gpu
05

MEASURED PERFORMANCE

METRICMEASUREDNOTES
StatusPreviewcoherent output via hybrid CPU-prefill + GPU-decode; expert-cache fast path in active development
Competitive bar (llama.cpp -cmoe)20 tok/s on 8 GB RTX 4060community-measured; the target this recipe is being tuned against

Expert-cache mechanics measured on hybrid-MoE (Cascade-2) 2026-07-20; 26B-A4B throughput pending.

Marked PREVIEW: decode-side expert-cache acceleration for this family is in active development; published throughput numbers will follow the same same-session A/B standard as every other recipe.