[Issue] Qwen3 Number of parameters displayed incorrectly

[hrsmanian]

1. I was recently finetuning Qwen3 1.7B params model for a use case and using LoRA. Noticed this output in the finetuning script

==((====))== Unsloth - 2x faster free finetuning | Num GPUs used = 1
\ /| Num examples = 22,948 | Num Epochs = 1 | Total steps = 11,474
O^O/ _/ \ Batch size per device = 1 | Gradient accumulation steps = 2
\ / Data Parallel GPUs = 1 | Total batch size (1 x 2 x 1) = 2
"-____-" Trainable parameters = 8,716,288/7,000,000,000 (0.12% trained)

It is surprising to see number of total parameters showing as 7B instead of 1.7B. Wondering why would that be? Llama 3B params are displayed correctly. Is this an issue with the code?

I am using the "unsloth/Qwen3-1.7B-bnb-4bit" model to train and using

unsloth==2025.4.7
unsloth-zoo==2025.5.8

Thanks

[Datta0]

Hey @hrsmanian thanks for reporting this issue. I have added a fix for the same in this PR
#2662

It'd be great if you can check if this solves it for you. I have checked from my end