vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

cuda : fix supports_op for norm

Browse source
This commit is contained in:
slaren 2024-10-31 17:49:51 +01:00
parent b135927ca4
commit d1faeca19d
1 changed files with 4 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ggml/src/ggml-cuda.cu
View file

@ -3107,18 +3107,20 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
} }
return false; return false;
} break; } break;
case GGML_OP_NORM:
case GGML_OP_RMS_NORM:
return ggml_is_contiguous(op->src[0]) && op->ne[0] % WARP_SIZE == 0;
break;
case GGML_OP_NONE: case GGML_OP_NONE:
case GGML_OP_RESHAPE: case GGML_OP_RESHAPE:
case GGML_OP_VIEW: case GGML_OP_VIEW:
case GGML_OP_PERMUTE: case GGML_OP_PERMUTE:
case GGML_OP_TRANSPOSE: case GGML_OP_TRANSPOSE:
case GGML_OP_NORM:
case GGML_OP_ADD: case GGML_OP_ADD:
case GGML_OP_ADD1: case GGML_OP_ADD1:
case GGML_OP_SUB: case GGML_OP_SUB:
case GGML_OP_MUL: case GGML_OP_MUL:
case GGML_OP_DIV: case GGML_OP_DIV:
case GGML_OP_RMS_NORM:
case GGML_OP_SCALE: case GGML_OP_SCALE:
case GGML_OP_SQR: case GGML_OP_SQR:
case GGML_OP_SQRT: case GGML_OP_SQRT: