nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 21 Packages Projects Releases Wiki Activity

support bf16 and quantized type (#20803)

Browse Source
This commit is contained in:
Neo Zhang
2026-03-22 22:06:27 +08:00
committed by GitHub
parent db9d8aa428
commit f40a80b4f3
1 changed files with 1 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-sycl/ggml-sycl.cpp
+1 -15
View File
@@ -4667,22 +4667,8 @@ static bool ggml_backend_sycl_device_supports_op(ggml_backend_dev_t dev, const g
if (a->ne[3] != b->ne[3]) { if (a->ne[3] != b->ne[3]) {
return false; return false;
} }
ggml_type a_type = a->type;
if (a_type == GGML_TYPE_IQ4_NL || a_type == GGML_TYPE_IQ4_XS ||
a_type == GGML_TYPE_IQ3_XXS || a_type == GGML_TYPE_IQ3_S ||
a_type == GGML_TYPE_IQ2_XXS || a_type == GGML_TYPE_IQ2_XS || a_type == GGML_TYPE_IQ2_S ||
a_type == GGML_TYPE_IQ1_S || a_type == GGML_TYPE_IQ1_M
) {
if (b->ne[1] == 1 && ggml_nrows(b) > 1) {
return false;
}
}
ggml_type src0_type = op->src[0]->type; ggml_type src0_type = op->src[0]->type;
if (src0_type == GGML_TYPE_BF16 ) {
// TODO: support GGML_TYPE_BF16
// FIXME: keep a list of supported types to avoid breaking the backend when a new type is added
return false;
}
// TODO: The configuration below needs more work to be supported with oneDNN // TODO: The configuration below needs more work to be supported with oneDNN
if (ggml_is_permuted(a) && !ggml_is_contiguous(a) && if (ggml_is_permuted(a) && !ggml_is_contiguous(a) &&