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

ggml : fix q4xx mat mul, increase ggml_aligned_malloc alignment (#10167)

Browse source
This commit is contained in:
Diego Devesa 2024-11-04 17:34:08 +01:00 • committed by GitHub
parent 9e0ecfb697
commit 401558b7ba
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 8 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

5
ggml/src/ggml-cpu.c
View file

@ -304,6 +304,7 @@ static const struct ggml_type_traits_cpu type_traits_cpu[GGML_TYPE_COUNT] = {
.nrows = 1,
},
[GGML_TYPE_Q8_0] = {
.from_float_to_mat = quantize_mat_q8_0,
.vec_dot = ggml_vec_dot_q8_0_q8_0,
.vec_dot_type = GGML_TYPE_Q8_0,
#if defined (__ARM_FEATURE_MATMUL_INT8)
@ -13692,9 +13693,7 @@ void ggml_cpu_init(void) {
uint16_t u16;
ggml_fp16_t fp16;
} u = {i};
// FIXME: this table is used in conversion functions outside of compute
// current code depends on ggml_init initializing this table
float f = ggml_table_f32_f16[i] = GGML_COMPUTE_FP16_TO_FP32(u.fp16);
float f = GGML_FP16_TO_FP32(u.fp16);
ggml_table_gelu_f16[i] = GGML_FP32_TO_FP16(ggml_gelu_f32(f));
ggml_table_gelu_quick_f16[i] = GGML_FP32_TO_FP16(ggml_gelu_quick_f32(f));
}