Chroma
is a 8.9 billion parameter rectified flow transformer capable of generating images from text descriptions.
Based on
FLUX.1 [schnell]
with heavy architectural modifications.
Quantized into GGUF format using a modified llama.cpp & city96's ComfyUI-GGUF/tools. Distillation layers are not quantized.
Also see
silveroxides'
Chroma GGUFs! (BF16, Q8_0, Q6_K, Q5_K_S, Q5_1, Q5_0, Q4_K_M, Q4_1, Q4_0, Q3_K_L)
Q*_M GGUFs are mixed quantizations with an aim at maximizing speed by selectively choosing the quantization of certain layers.
-
Q8_M focuses on Q8_0 quantization of weights for performance, mixed with Q6_K on less heavy layers.