Add enum llama_ftype, sync ggml_type to model files (#709)

1 files changed, 16 insertions, 19 deletions

diff --git a/ggml.c b/ggml.c
index 897b67d..31947c4 100644
--- a/ggml.c
+++ b/ggml.c
@@ -2560,29 +2560,26 @@ inline static void ggml_vec_norm_inv_f32(const int n, float * s, const float * x
 //
 
 static const int GGML_BLCK_SIZE[GGML_TYPE_COUNT] = {
-    QK,
-    QK,
-    1,
-    1,
-    1,
-    1,
-    1,
+    [GGML_TYPE_F32]  = 1,
+    [GGML_TYPE_F16]  = 1,
+    [GGML_TYPE_Q4_0] = QK,
+    [GGML_TYPE_Q4_1] = QK,
+    [GGML_TYPE_I8]   = 1,
+    [GGML_TYPE_I16]  = 1,
+    [GGML_TYPE_I32]  = 1,
 };
-
-static_assert(GGML_TYPE_COUNT == 7, "GGML_TYPE_COUNT != 5");
+static_assert(GGML_TYPE_COUNT == 7, "GGML_BLCK_SIZE is outdated");
 
 static const size_t GGML_TYPE_SIZE[GGML_TYPE_COUNT] = {
-    sizeof(block_q4_0),
-    sizeof(block_q4_1),
-    sizeof(int8_t ),
-    sizeof(int16_t),
-    sizeof(int32_t),
-    sizeof(ggml_fp16_t),
-    sizeof(float  ),
+    [GGML_TYPE_F32]  = sizeof(float),
+    [GGML_TYPE_F16]  = sizeof(ggml_fp16_t),
+    [GGML_TYPE_Q4_0] = sizeof(block_q4_0),
+    [GGML_TYPE_Q4_1] = sizeof(block_q4_1),
+    [GGML_TYPE_I8]   = sizeof(int8_t),
+    [GGML_TYPE_I16]  = sizeof(int16_t),
+    [GGML_TYPE_I32]  = sizeof(int32_t),
 };
-
-// don't forget to update the array above when adding new types
-static_assert(GGML_TYPE_COUNT == 7, "GGML_TYPE_COUNT != 5");
+static_assert(GGML_TYPE_COUNT == 7, "GGML_TYPE_SIZE is outdated");
 
 static const char * GGML_OP_LABEL[GGML_OP_COUNT] = {
     "NONE",