mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-04-24 00:19:19 +00:00
WIP
This commit is contained in:
File diff suppressed because it is too large
Load Diff
3932
ggml/src/ggml-cuda/mmq_id_common.cuh
Normal file
3932
ggml/src/ggml-cuda/mmq_id_common.cuh
Normal file
File diff suppressed because it is too large
Load Diff
22
ggml/src/ggml-cuda/mmq_id_kernels.cu
Normal file
22
ggml/src/ggml-cuda/mmq_id_kernels.cu
Normal file
@@ -0,0 +1,22 @@
|
||||
#include "mmq_id_common.cuh"
|
||||
|
||||
DECL_MMQ_CASE(GGML_TYPE_Q4_0);
|
||||
DECL_MMQ_CASE(GGML_TYPE_Q4_1);
|
||||
DECL_MMQ_CASE(GGML_TYPE_Q5_0);
|
||||
DECL_MMQ_CASE(GGML_TYPE_Q5_1);
|
||||
DECL_MMQ_CASE(GGML_TYPE_Q8_0);
|
||||
DECL_MMQ_CASE(GGML_TYPE_MXFP4);
|
||||
DECL_MMQ_CASE(GGML_TYPE_Q2_K);
|
||||
DECL_MMQ_CASE(GGML_TYPE_Q3_K);
|
||||
DECL_MMQ_CASE(GGML_TYPE_Q4_K);
|
||||
DECL_MMQ_CASE(GGML_TYPE_Q5_K);
|
||||
DECL_MMQ_CASE(GGML_TYPE_Q6_K);
|
||||
DECL_MMQ_CASE(GGML_TYPE_IQ2_XXS);
|
||||
DECL_MMQ_CASE(GGML_TYPE_IQ2_XS);
|
||||
DECL_MMQ_CASE(GGML_TYPE_IQ2_S);
|
||||
DECL_MMQ_CASE(GGML_TYPE_IQ3_XXS);
|
||||
DECL_MMQ_CASE(GGML_TYPE_IQ3_S);
|
||||
DECL_MMQ_CASE(GGML_TYPE_IQ1_S);
|
||||
DECL_MMQ_CASE(GGML_TYPE_IQ4_NL);
|
||||
DECL_MMQ_CASE(GGML_TYPE_IQ4_XS);
|
||||
|
||||
Reference in New Issue
Block a user