vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-08-08 07:55:25 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

ggml-webgpu: add support for f16 repeat (#26307)

Browse source
This commit is contained in:
Masashi Yoshimura 2026-08-02 15:28:31 +09:00 • committed by GitHub
parent 11924d4c17
commit 7a2db1a0cf
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 10 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp
View file

@ -2774,6 +2774,10 @@ class ggml_webgpu_shader_lib {
defines.push_back("TYPE_F32");
variant += "_f32";
break;
case GGML_TYPE_F16:
defines.push_back("TYPE_F16");
variant += "_f16";
break;
case GGML_TYPE_I32:
defines.push_back("TYPE_I32");
variant += "_i32";

3
ggml/src/ggml-webgpu/ggml-webgpu.cpp
View file

@ -4290,7 +4290,8 @@ static bool ggml_backend_webgpu_device_supports_op(ggml_backend_dev_t dev, const
supports_op = (src0->type == GGML_TYPE_F32 || src0->type == GGML_TYPE_I32);
break;
case GGML_OP_REPEAT:
supports_op = (src0->type == GGML_TYPE_F32 || src0->type == GGML_TYPE_I32 || src0->type == GGML_TYPE_I16);
supports_op = (src0->type == GGML_TYPE_F32 || src0->type == GGML_TYPE_F16 || src0->type == GGML_TYPE_I32 ||
src0->type == GGML_TYPE_I16);
break;
case GGML_OP_CPY:
case GGML_OP_CONT:

3
ggml/src/ggml-webgpu/wgsl-shaders/repeat.wgsl
View file

@ -27,6 +27,9 @@ struct Params {
#ifdef TYPE_I32
#define DataType i32
#endif
#ifdef TYPE_F16
#define DataType f16
#endif
#ifdef TYPE_I16
// same size (16-bit) is sufficient for repeat
#define DataType f16

1
tests/test-backend-ops.cpp
View file

@ -8511,6 +8511,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_repeat(GGML_TYPE_F32, {10, 5, 4, ne3}, {1, 2, 1, 1}));
test_cases.emplace_back(new test_repeat(GGML_TYPE_F32, {10, 5, 4, ne3}, {1, 1, 2, 1}));
test_cases.emplace_back(new test_repeat(GGML_TYPE_F32, {10, 5, 4, ne3}, {1, 1, 1, 2}));
test_cases.emplace_back(new test_repeat(GGML_TYPE_F16, {10, 5, 4, ne3}, {2, 1, 1, 1}));
test_cases.emplace_back(new test_repeat(GGML_TYPE_I32, {10, 5, 4, ne3}, {2, 1, 1, 1}));
test_cases.emplace_back(new test_repeat(GGML_TYPE_I16, {10, 5, 4, ne3}, {1, 1, 1, 2}));
test_cases.emplace_back(new test_repeat(GGML_TYPE_BF16, {10, 5, 4, ne3}, {2, 1, 1, 1}));