GCMP: same accuracy as LoRA r=32 on Qwen3.8-27B with 55× fewer parameters (11 MB vs 320 MB per adapter)

🌐 链接: https://www.reddit.com/r/LocalLLaMA/comments/1xy5eby/gcmp_same_accuracy_as_lora_r32_on_qwen3827b_with/
🏷️ 分组: Reddit|AI·自托管·VPS
🕒 时间: 2026-11-17 07:56:50 From the LocalLLaMA community on Reddit: GCMP: same accuracy as LoRA r=32 on Qwen3.8-27B with 55× fewer parameters (11 MB vs 320…