LFM2.5MoE
LFM2.5 8B-A1B
MoE compact avec seulement 1B actif, pense pour debit et tool use.
VRAM min 4.5 GoParams 8.47B
ChoisirVRAM, quant, benchmarks, tok/s. Direct.
MoE compact avec seulement 1B actif, pense pour debit et tool use.
Variante thinking sous 2B pour raisonner localement sans grosse machine.
Taille centrale Gemma pour chat local, vision et GPU grand public confortables.
MoE moderne avec bon ratio qualite/calcul pour agents locaux exigeants.