furyhawk/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-07-21 10:15:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: only support F32/F16 for GGML_OP_REPEAT (#24533)

Browse Source
This commit is contained in:
leonardHONG
2026-06-15 09:11:00 +03:00
committed by GitHub
parent 6e9007ae61
commit 3686e9d643
2 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/ggml-cuda.cu
+2 -1
View File
@@ -5337,8 +5337,9 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
} break;
case GGML_OP_REPEAT:
{
// the CUDA REPEAT path only implements F32/F16; other types assert at runtime
ggml_type src0_type = op->src[0]->type;
return src0_type != GGML_TYPE_I32 && src0_type != GGML_TYPE_I16;
return src0_type == GGML_TYPE_F32 || src0_type == GGML_TYPE_F16;
} break;
case GGML_OP_REPEAT_BACK:
return op->type == GGML_TYPE_F32 && (op->src[0]->ne[2]*op->src[0]->ne[3]) <= (1 << 15);
tests/test-backend-ops.cpp
+1
View File
@@ -8086,6 +8086,7 @@ 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, 1, 1, 2}));
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}));
}
for (bool view : {false, true}) {