2025-12-13 17:50:47 +01:00
We have one horrible disjuncture, between layers 6 → 2. I have one more hypothesis: A little bit of fine-tuning on those two layers is all we really need. Fine-tuned RYS models dominate the Leaderboard. I suspect this junction is exactly what the fine-tuning fixes. And there’s a great reason to do this: this method does not use extra VRAM! For all these experiments, I duplicated layers via pointers; the layers are repeated without using more GPU memory. Of course, we do need more compute and more KV cache, but that’s a small price to pay for a verifiably better model. We can just ‘fix’ an actual copies of layers 2 and 6, and repeat layers 3-4-5 as virtual copies. If we fine-tune all layer, we turn virtual copies into real copies, and use up more VRAM.
，详情可参考有道翻译下载
Экс-вокалистка коллектива "Демо" 45 лет появилась в магазине в прозрачном наряде20:46。业内人士推荐Facebook BM账号,Facebook企业管理,Facebook商务账号作为进阶阅读
00:52, 9 марта 2026Мир。关于这个话题，有道翻译提供了深入分析
3 апреля 2026, 05:16 Международная панорама
中东各国开始筹备伊朗即将到来的末日危机 19:51