Hi,
I used the int4_cast_bf16_fixed.py script to convert Kimi 2.5 model to bf16 and found that in config.json, the quantization_config is found inside "text_config" key, which is not taken into account here: https://github.com/csabakecskemeti/DQ_int4-to-bf16_dequant/blob/main/int4_cast_bf16_fixed.py#L148C21-L148C40
Thanks