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

CUDA: fuse rms_norm + mul + rope (+ view + set_rows) (#26767)

Browse Source 
* CUDA: fuse rms_norm + mul + rope (+ view + set_rows)

* tests: add broadcast weight case to rms_norm_mul_rope

* CUDA: check memory ranges before rms_norm rope fusion

* CUDA: check memory ranges in rope set_rows fusion
This commit is contained in:
Rafail Giavrimis
2026-08-09 00:32:37 +08:00
committed by GitHub
parent 18f7ad7fc9
commit 687e778927
4 changed files with 344 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/rope.cuh
+2
View File
@@ -7,3 +7,5 @@ void ggml_cuda_op_rope(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
void ggml_cuda_op_rope_back(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
void ggml_cuda_op_rope_fused(ggml_backend_cuda_context & ctx, ggml_tensor * dst, ggml_tensor * set_rows);
void ggml_cuda_op_rms_norm_mul_rope_fused(ggml_backend_cuda_context & ctx, ggml_tensor * rms_norm, ggml_tensor * mul, ggml_tensor * rope, ggml_tensor * set_rows);