Diffusers Flux-2 (FLUX.2-dev-bnb-4bit) on CUDA hits OutOfMemoryError during inference
AI Impact Summary
The snippet loads a Flux-2 diffusion pipeline from diffusers using a 4-bit FLUX.2 model and a separate Mistral-3 text encoder, then attempts to generate an image on CUDA with CPU offload. The run terminates with an OutOfMemoryError during the inference flow, indicating memory pressure from the combined transformer and diffusion components. The repeated Flax deprecation warnings in the logs suggest cross-version compatibility issues; while not a direct action item, they foreshadow potential maintenance constraints as Diffusers evolves. Businesswise, this memory failure directly disrupts automated image generation workflows that rely on Flux-2, delaying content creation and any downstream processes that depend on timely image outputs.
Affected Systems
- Date
- Date not specified
- Change type
- capability
- Severity
- info