@@ -89,16 +89,17 @@ struct log_scope {
89
89
static void test_llama_kv_cache_unified_constructor () {
90
90
auto model = _make_model ();
91
91
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
102
103
);
103
104
}
104
105
@@ -113,11 +114,11 @@ static void test_llama_kv_cache_unified_single_seq() {
113
114
/* v_trans */ false ,
114
115
/* offload */ false ,
115
116
/* kv_size */ 10 ,
117
+ /* n_seq_max */ 1 ,
116
118
/* padding */ 10 ,
117
119
/* n_swa */ 0 ,
118
120
/* swa_type */ LLAMA_SWA_TYPE_NONE
119
121
);
120
- GGML_ASSERT (cache.get_used_cells () == 0 );
121
122
122
123
// Create the micro batch with a single 3-token sequence
123
124
//
@@ -155,11 +156,12 @@ static void test_llama_kv_cache_unified_single_seq() {
155
156
static void test_llama_kv_cache_recurrent_constructor () {
156
157
auto model = _make_model (LLM_ARCH_MAMBA);
157
158
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
163
165
);
164
166
}
165
167
0 commit comments