ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-23 22:54:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

WIP

Browse Source
This commit is contained in:
Iwan Kawrakow
2025-08-24 20:03:55 +03:00
parent 916733144c
commit 26c0dfdbfa
20 changed files with 89 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
ggml/src/ggml-cuda/mmq_id_common.cuh
View File

@@ -3746,9 +3746,6 @@ static void launch_mul_mat_q_id(ggml_backend_cuda_context & ctx, const mmq_args_
const int ntzw = args.nchannels_y * args.nsamples_y;
const dim3 block_nums_xy_tiling(nty, ntx, ntzw);
if (args.nchannels_y % args.nchannels_x) {
printf("Oops: args.nchannels_y = %d, args.nchannels_x = %d\n", args.nchannels_y, args.nchannels_x);
}
GGML_ASSERT(args.nchannels_y % args.nchannels_x == 0);
GGML_ASSERT(args.nsamples_y % args.nsamples_x == 0);
const int channel_ratio = args.nchannels_y / args.nchannels_x;

22
ggml/src/ggml-cuda/mmq_id_kernels.cu
View File

@@ -1,22 +0,0 @@
#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);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-iq1_s_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_IQ1_S);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_s_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_IQ2_S);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_xs_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_IQ2_XS);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_xxs_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_IQ2_XXS);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-iq3_s_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_IQ3_S);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-iq3_xxs_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_IQ3_XXS);

4
ggml/src/ggml-cuda/template-instances/mmq-instance-iq4_nl_id.cu Normal file
View File

@@ -0,0 +1,4 @@
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_IQ4_NL);
DECL_MMQ_CASE(GGML_TYPE_MXFP4);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-iq4_xs_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_IQ4_XS);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-q2_k_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_Q2_K);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-q3_k_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_Q3_K);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-q4_0_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_Q4_0);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-q4_1_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_Q4_1);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-q4_k_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_Q4_K);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-q5_0_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_Q5_0);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-q5_1_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_Q5_1);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-q5_k_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_Q5_K);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-q6_k_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_Q6_K);

5
ggml/src/ggml-cuda/template-instances/mmq-instance-q8_0_id.cu Normal file
View File

@@ -0,0 +1,5 @@
// This file has been autogenerated by generate_cu_files.py, do not edit manually.
#include "../mmq_id_common.cuh"
DECL_MMQ_CASE(GGML_TYPE_Q8_0);