Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

CUDA: topk-moe: add optional parameter for gpt-oss (#16649)

Browse Source
This commit is contained in:
Aman Gupta
2025-10-21 22:40:38 +08:00
committed by GitHub
parent 51d1a8c997
commit 03792ad936
4 changed files with 154 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/topk-moe.cuh
+4 -3
View File
@@ -6,9 +6,10 @@
void ggml_cuda_op_topk_moe(ggml_backend_cuda_context & ctx,
const ggml_tensor * logits,
ggml_tensor * weights,
ggml_tensor * top_k,
const bool with_norm);
ggml_tensor * ids,
const bool with_norm,
const bool delayed_softmax = false);
bool ggml_cuda_should_use_topk_moe(const ggml_tensor * softmax, const ggml_tensor * weights);
std::initializer_list<enum ggml_op> ggml_cuda_topk_moe_ops(bool with_norm);
std::initializer_list<enum ggml_op> ggml_cuda_topk_moe_ops(bool with_norm, bool delayed_softmax = false);