works in comfyUI but not fully supported yet, vram/ram spikes. most likely fixed within a day
Works with normal klein 9b loras
allowed txt attn to quant
flux2-klein-9b-kv-nvfp4.safetensors
txt attn kept at bf16
flux2-klein-9b-kv-nvfp4_txtattnBF16.safetensors