K2-Horizon-MoVA-36B-A4B with multi-head latent attention (TransMLA, 2.3x smaller KV cache) and SeedLM seed weights, vs oQe.
-
txgsync/K2-Horizon-MoVA-36B-A4B-MLA-NanoSeedLM-P4
Text Generation • 12B • Updated • 78 -
txgsync/K2-Horizon-MoVA-36B-A4B-MLA-NanoSeedLM-P3
Text Generation • 10B • Updated • 78 -
txgsync/K2-Horizon-MoVA-36B-A4B-MLA
Text Generation • 38B • Updated • 536 • 1 -
IFM/K2-Horizon-MoVA-36B-A4B
Text Generation • 37B • Updated • 17.6k • • 414