FLUX.2-klein-base-4B — OpenVINO INT8
OpenVINO IR (INT8) conversion of black-forest-labs/FLUX.2-klein-base-4B (4B rectified-flow transformer + text encoder + VAE). Standard diffusers layout verified before conversion.
Why OpenVINO?
- Runs on any CPU with no CUDA/NVIDIA dependency, plus GPU acceleration where available (AMD, Nvidia, Intel) — no vendor-locked passes used
- Full pipeline is ~8GB, deployable on 8GB VRAM cards
Deployment note (8GB cards)
Transformer (~3.9GB INT8) runs on GPU; the text encoder runs once per prompt
and can sit on CPU via AUTO/explicit device placement, leaving comfortable
headroom on 8GB cards.
Conversion
| Parameter | Value |
|---|---|
| Tool | optimum-cli export openvino --weight-format int8 --all-layers (device-neutral) |
| Precision | INT8 |
| Source | black-forest-labs/FLUX.2-klein-base-4B |
| Verified | transformer, text_encoder, vae_encoder, vae_decoder all load in ov.Core on Kaggle GPU |
Usage (any device)
import openvino as ov
core = ov.Core()
compiled = core.compile_model(core.read_model("transformer/openvino_model.xml"), "AUTO")