AIE

Quantization: Performance & Quality

FLUX.2(dev) on Blackwell | Quality preserved at every precision level

BF16

NVFP4

DIY with TensorRT-LLM - Visual Gen

Run a pre-quantized ckpt from HF

# With NVFP4 quantization

python visual_gen_flux.py --model_path black-forest-labs/FLUX.2-dev --prompt "A cat" --linear_type trtllm-nvfp4

black-forest-labs/FLUX.2-dev-NVFP4

AI Engineer EUROPE