3 comments
Very cool stuff. Love the focus on CPU-first.<p>Would also love to see some throughput numbers on basic VM setup.<p>Edit: there are some latency numbers in the paper <a href="https://arxiv.org/pdf/2507.18546" rel="nofollow">https://arxiv.org/pdf/2507.18546</a>
Zero-shot encoder models are so cool. I'll definitely be checking this out.<p>If you're looking for a zero-shot classifier, tasksource is in a similar vein.<p><a href="https://huggingface.co/tasksource/ModernBERT-large-nli" rel="nofollow">https://huggingface.co/tasksource/ModernBERT-large-nli</a>
There is another version at:<p><a href="https://github.com/urchade/GLiNER" rel="nofollow">https://github.com/urchade/GLiNER</a><p>Looks like it’s still being maintained too?