HuggingFaceTB

/

SmolVLM2-2.2B-Instruct

like 59

Follow

Hugging Face TB Research 1.25k

Hello 👋
When trying to reproduce the examples from the model card, I got an error when calling model.generate(...):
RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (CUDABFloat16Type) should be the same

I fixed it by adding param dtype=torch.bfloat16 when moving the processed inputs to CUDA (this is included in the last example "Multi-image Interleaved Inference", but not in the first two)
This should avoid mismatched tensor types between model weights and inputs.

Cheers,
Michal