ggml-org/Kev-0.8B-GGUF
Adapter vs jaredpalmer/kev-0.8b. jaredpalmer/kev-0.8b is a LoRA adapter, not full weights: rank 16, on down_proj, gate_proj, in_proj_a, in_proj_b, in_proj_qkv, in_proj_z, k_proj, o_proj… in 24 layers (186 adapted matrices). It modifies ggml-org/Kev-0.8B-GGUF at load time.