LLM-Inference/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-08-12 22:29:39 +04:00
Code Issues Packages Projects Releases Wiki Activity

Allow K to be f32 in ggml_cuda_op_indexer_topk

Browse Source
This commit is contained in:
Kawrakow
2026-08-02 16:24:04 +00:00
parent c7dfe6f08b
commit 6b1b43fe9d
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/indexer_topk.cu
+2 -1
View File
@@ -69,7 +69,8 @@ void ggml_cuda_op_indexer_topk(ggml_backend_cuda_context & ctx, ggml_tensor * ds
int n_top_k = dst->ne[0];
int n_kv = k->ne[1];
int n_head = q->ne[1];
GGML_ASSERT(k->type == GGML_TYPE_F16 || ggml_is_quantized(k->type));
//if (k->type != GGML_TYPE_F16 && !ggml_is_quantized(k->type)) printf("%s: K is %s?\n", __func__, ggml_type_name(k->type));
GGML_ASSERT(k->type == GGML_TYPE_F16 || k->type == GGML_TYPE_F32 || ggml_is_quantized(k->type));
GGML_ASSERT(k->ne[2] == 1 || k->ne[3] == 1);
GGML_ASSERT(k->ne[1] > n_top_k);
GGML_ASSERT(k->ne[1] == m->ne[0]);