Inicio Explorar Axuda
Iniciar sesión
cturan
/
llama.cpp
réplica de https://github.com/cturan/llama.cpp
1
0
Fork 0
Ficheiros Incidencias 0 Wiki
Explorar o código

cuda : get_row_rounding F32 (#4095)

* Fix #4017

* Update ggml-cuda.cu

Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>

* Update ggml-cuda.cu

Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>

---------

Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>
Andrew Godfrey %!s(int64=2) %!d(string=hai) anos
pai
4f447a4833
achega
b83e149ec6
Modificáronse 1 ficheiros con 2 adicións e 0 borrados
Unificar vista Mostrar estatísticas de Diff
  1. 2 0
      ggml-cuda.cu

+ 2 - 0
ggml-cuda.cu
Ver ficheiro 

@@ -6356,6 +6356,7 @@ static int64_t get_row_rounding(ggml_type type) {
 
         case GGML_TYPE_Q8_0:
 
         case GGML_TYPE_Q8_0:
 
             return max_compute_capability >= CC_RDNA2 ? 128 : 64;
 
             return max_compute_capability >= CC_RDNA2 ? 128 : 64;
 
         case GGML_TYPE_F16:
 
         case GGML_TYPE_F16:
 
+        case GGML_TYPE_F32:
 
             return 1;
 
             return 1;
 
         case GGML_TYPE_Q2_K:
 
         case GGML_TYPE_Q2_K:
 
             return max_compute_capability >= CC_RDNA2 ? 128 : 32;
 
             return max_compute_capability >= CC_RDNA2 ? 128 : 32;
 
@@ -6378,6 +6379,7 @@ static int64_t get_row_rounding(ggml_type type) {
 
         case GGML_TYPE_Q8_0:
 
         case GGML_TYPE_Q8_0:
 
             return 64;
 
             return 64;
 
         case GGML_TYPE_F16:
 
         case GGML_TYPE_F16:
 
+        case GGML_TYPE_F32:
 
             return 1;
 
             return 1;
 
         case GGML_TYPE_Q2_K:
 
         case GGML_TYPE_Q2_K:
 
         case GGML_TYPE_Q3_K:
 
         case GGML_TYPE_Q3_K: