feat(wip): Use type_k/type_v for hybrid cache types

Branch: Mamba2SSD

Signed-off-by: Gabe Goodhart <[email protected]>

Author: gabe-l-hart

src/llama-model.cpp

@@ -6786,8 +6786,8 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
                         /* attn_n_pad        */ 1,
                         /* attn_n_swa        */ hparams.n_swa,
                         /* attn_swa_type     */ hparams.swa_type,
-                        /* recurrent_type_k  */ GGML_TYPE_F32,
-                        /* recurrent_type_v  */ GGML_TYPE_F32,
+                        /* recurrent_type_r  */ params.type_k,
+                        /* recurrent_type_s  */ params.type_v,
                         /* recurrent_kv_size */ std::max((uint32_t) 1, cparams.n_seq_max),
                         /* n_seq_max         */ cparams.n_seq_max,
                         /* offload           */ cparams.offload_kqv,