Optimizing speed & quality on Qwen3.6 27b
Signal
35
Hype
15
En 3 lignesUtilisateur optimise l'inférence de Qwen 3.6 27B sur llama.cpp avec 40GB VRAM (RTX 2060 Super + 2x RTX 5060 Ti). Atteint 300-500 tok/s en traitement de prompt et 22-30 tok/s en génération à fenêtre de contexte 100k. Demande si configuration est optimale ou si améliorations possibles.Lire la source
Ton avis ?
Résumé généré par Claude — vérifié par l'humain