Fine-tuning NuExtract3 (LoRA) on 624 train / 44 eval books from this pilot corpus markedly improved precision (0.61→0.76), with recall essentially unchanged (0.76→0.75) vs. the zero-shot base model. Next: scale up the training set.Code: https://github.com/cboulanger/finetune-nuextract-tocTry it: https://huggingface.co/spaces/cmboulanger/nuextract3-toc-demo (3/3)#MachineLearning #OpenSource #AI
Related
📰 Daily Bulletins - 10/11/2026📉 Collapse : 🟠 4/5🌍 Planet : 🟠 4/5🛡️ War in Europe : 🟠 4/5🤖 AGI risk : 🟠 4/5🔗 https://etik...
📰 Daily Bulletins - 10/11/2026📉 Collapse : 🟠 4/5🌍 Planet : 🟠 4/5🛡️ War in Europe : 🟠 4/5🤖 AGI risk : 🟠 4/5🔗 https://etik.soulignac.eu/bulletin/mardi?lang=en#collapse #climate #war ...
Companies such as Google used to be our portals to the rest of the web. Can the internet as we know it survive when A.I....
Companies such as Google used to be our portals to the rest of the web. Can the internet as we know it survive when A.I. answers all of our questions? #culture #search #aihttps://w...
The authors show that transformer models trained with discrete diffusion and decoded by guided beam search can recover t...
The authors show that transformer models trained with discrete diffusion and decoded by guided beam search can recover the prime factors of semiprime integers directly from data.So...