@@ -89,16 +89,17 @@ struct log_scope {
8989static void test_llama_kv_cache_unified_constructor () {
9090 auto model = _make_model ();
9191 llama_kv_cache_unified cache (
92- /* model */ *model,
93- /* filter */ nullptr ,
94- /* type_k */ GGML_TYPE_F32,
95- /* type_v */ GGML_TYPE_F16,
96- /* v_trans */ false ,
97- /* offload */ false ,
98- /* kv_size */ 10 ,
99- /* padding */ 10 ,
100- /* n_swa */ 0 ,
101- /* swa_type */ LLAMA_SWA_TYPE_NONE
92+ /* model */ *model,
93+ /* filter */ nullptr ,
94+ /* type_k */ GGML_TYPE_F32,
95+ /* type_v */ GGML_TYPE_F16,
96+ /* v_trans */ false ,
97+ /* offload */ false ,
98+ /* kv_size */ 10 ,
99+ /* n_seq_max */ 1 ,
100+ /* padding */ 10 ,
101+ /* n_swa */ 0 ,
102+ /* swa_type */ LLAMA_SWA_TYPE_NONE
102103 );
103104}
104105
@@ -113,11 +114,11 @@ static void test_llama_kv_cache_unified_single_seq() {
113114 /* v_trans */ false ,
114115 /* offload */ false ,
115116 /* kv_size */ 10 ,
117+ /* n_seq_max */ 1 ,
116118 /* padding */ 10 ,
117119 /* n_swa */ 0 ,
118120 /* swa_type */ LLAMA_SWA_TYPE_NONE
119121 );
120- GGML_ASSERT (cache.get_used_cells () == 0 );
121122
122123 // Create the micro batch with a single 3-token sequence
123124 //
@@ -155,11 +156,12 @@ static void test_llama_kv_cache_unified_single_seq() {
155156static void test_llama_kv_cache_recurrent_constructor () {
156157 auto model = _make_model (LLM_ARCH_MAMBA);
157158 llama_kv_cache_recurrent cache (
158- /* model */ *model,
159- /* type_k */ GGML_TYPE_F32,
160- /* type_v */ GGML_TYPE_F16,
161- /* offload */ false ,
162- /* kv_size */ 10
159+ /* model */ *model,
160+ /* type_k */ GGML_TYPE_F32,
161+ /* type_v */ GGML_TYPE_F16,
162+ /* offload */ false ,
163+ /* kv_size */ 10 ,
164+ /* n_seq_max */ 1
163165 );
164166}
165167
0 commit comments