Beware: doesn't work on compute capability 8.6, only on 8.9 and newer

#1
by DrRos - opened

Subj. Do not download if your GPU is not supported.
Value error, The quantization method modelopt is not supported for the current GPU. Minimum capability: 89. Current capability: 86.

Yep, this wasted my time (and bandwidth) too.

NVIDIA, will the bf16 model work OK, or does it have this modelopt thing too? Can you release versions that don't require that, for the other GPUs that are still in support?

You have lost nothing, the model without system prompt reports it's a 16B Qwen3 and has failed to count the 'r's in the word strarwberrry 5 times (never got the correct result)

Sign up or log in to comment