EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

vulkan: Allow up to 4096 elements for mul_mat_id row_ids (#13326)

Browse Source 
This assert fired running Qwen_Qwen3-30B-A3B-Q2_K.gguf:

GGML_ASSERT(nei0 * nei1 <= 3072);

The tensor is 8 x 512. Increase this array size to accommodate.
This commit is contained in:
Jeff Bolz
2025-05-09 02:23:41 -05:00
committed by GitHub
parent d9c4accaff
commit 02115dcd9a
4 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp
View File

@@ -101,7 +101,7 @@ shared FLOAT_TYPE_VEC2 buf_b_ds[BN];
#define LOAD_VEC_B 4
#ifdef MUL_MAT_ID
shared u16vec2 row_ids[3072];
shared u16vec2 row_ids[4096];
#endif // MUL_MAT_ID
#define NUM_WARPS (BLOCK_SIZE / WARP)