vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

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

Browse source
This commit is contained in:
Stephan Walter 2023-04-11 15:03:51 +00:00 • committed by GitHub
parent 2663d2c678
commit 3e6e70d8e8
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
5 changed files with 74 additions and 57 deletions
Download patch file Download diff file Expand all files Collapse all files

35
ggml.c
View file

@ -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static_assert(GGML_TYPE_COUNT == 7, "GGML_BLCK_SIZE is outdated");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static_assert(GGML_TYPE_COUNT == 7, "GGML_TYPE_SIZE is outdated");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",