sdgoij/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-05-14 04:54:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

wip

Browse Source
This commit is contained in:
Georgi Gerganov
2026-02-10 20:50:47 +02:00
parent 1c312dc758
commit e480e383fd
2 changed files with 5 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-cuda/ggml-cuda.cu
View File

@@ -4542,6 +4542,8 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
case GGML_UNARY_OP_CEIL:
case GGML_UNARY_OP_ROUND:
case GGML_UNARY_OP_TRUNC:
// TODO: should become:
//return ggml_is_contiguous_rows(op->src[0]);
return ggml_is_contiguous(op->src[0]);
default:
return false;

3
src/models/qwen3next.cpp
View File

@@ -504,6 +504,9 @@ ggml_tensor * llm_build_qwen3next::build_layer_attn(
cb(Qcur, "Qcur", il);
cb(gate, "gate", il);
// TODO: CUDA is missing non-contiguous unary ops. when implemented: remove this cont
gate = ggml_cont_2d(ctx0, gate, n_embd_head * n_head, n_tokens);
gate = ggml_sigmoid(ctx0, gate);
cb(gate, "gate_sigmoid", il);