Flux Klein 4B mixed precision text encoder

[SFrav]

Is the qwen_3_4b_fp4_flux2.safetensors file working in the recent update? I se that all the examples use the qwen_3_4b.safetensors file. When testing I get the following error: "[ERROR] stable-diffusion.cpp:781 - load tensors from model loader failed [INFO ] main.cpp:690 - new_sd_ctx_t failed"

P.S. Thank you for making this available. It would be impossible for me to do cpu inference otherwise.

[stduhpf]

Fp4 isn't supported by ggml

[stduhpf]

@SFrav You can use GGUF quants, for example, Q6_K_XL is about the same size as the fp4, and I beleive it should work a bit better.