This is one of those situations where I wish I had B200s to recompute these scores, along with an older model like Qwen3 32B
Because these quants really seem as good as Qwen3 32B on these benchmarks...
That's quite a gap!
Thanks unsloth for doing these evals
I wonder how this correlates with EAR. This I might be able to run, but ultimately, task accuracy is what matters