Matrix multiplication issue

[pulkitmehtaworkmetacube]

We are doing following :

1. Pick 4 bit quantized model https://huggingface.co/unsloth/Llama-3.1-Nemotron-70B-Instruct-bnb-4bit
2. Use that to finetune adapterH or bottleneck adapter.

We reduced bottleck_dimension to 64 instead of 256 .
It loads base model but during finetuning of adapter , it is giving following error .

mat1 and mat2 shapes cannot be multiplied (1024x8192 and 1x117440512)

Please suggest