CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-28 08:31:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

CANN: Mask unsupported TRANSPOSE_1D operator (#15733)

Browse Source 
CANN currently does not support kernels larger than 255.
This change disables such cases.
This commit is contained in:
hipudding
2025-09-03 14:08:22 +08:00
committed by GitHub
parent 8a2234ea0c
commit f6da8cb86a
2 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-cann/aclnn_ops.cpp
View File

@@ -2803,8 +2803,6 @@ void ggml_cann_conv_transpose_1d(ggml_backend_cann_context& ctx, ggml_tensor* ds
aclIntArray *padding = aclCreateIntArray(paddingVal, 1); aclIntArray *padding = aclCreateIntArray(paddingVal, 1);
int64_t dilationVal[] = {1}; int64_t dilationVal[] = {1};
aclIntArray *dilation = aclCreateIntArray(dilationVal, 1); aclIntArray *dilation = aclCreateIntArray(dilationVal, 1);
bool transposed = true;
int64_t groups = 1;
int8_t cubeMathType = 0; int8_t cubeMathType = 0;
#ifdef ASCEND_310P #ifdef ASCEND_310P
@@ -2812,7 +2810,7 @@ void ggml_cann_conv_transpose_1d(ggml_backend_cann_context& ctx, ggml_tensor* ds
#endif #endif
GGML_CANN_CALL_ACLNN_OP(ctx, Convolution, acl_input, acl_weight, nullptr, stride, GGML_CANN_CALL_ACLNN_OP(ctx, Convolution, acl_input, acl_weight, nullptr, stride,
padding, dilation, transposed, padding, groups, acl_dst, cubeMathType); padding, dilation, true, padding, 1, acl_dst, cubeMathType);
ggml_cann_release_resources(ctx, acl_weight, acl_dst, stride, padding, dilation); ggml_cann_release_resources(ctx, acl_weight, acl_dst, stride, padding, dilation);
} }

4
ggml/src/ggml-cann/ggml-cann.cpp
View File

@@ -2479,12 +2479,14 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
case GGML_OP_ARGMAX: case GGML_OP_ARGMAX:
case GGML_OP_COS: case GGML_OP_COS:
case GGML_OP_SIN: case GGML_OP_SIN:
case GGML_OP_CONV_TRANSPOSE_1D:
case GGML_OP_LOG: case GGML_OP_LOG:
case GGML_OP_MEAN: case GGML_OP_MEAN:
case GGML_OP_PAD_REFLECT_1D: case GGML_OP_PAD_REFLECT_1D:
case GGML_OP_COUNT_EQUAL: case GGML_OP_COUNT_EQUAL:
return true; return true;
case GGML_OP_CONV_TRANSPOSE_1D:
// TODO: ((weightL - 1) * dilationW - padLeft)=1336 should not be larger than 255.
return (op->src[0]->ne[0] - 1) <= 255;
case GGML_OP_SCALE: case GGML_OP_SCALE:
float bias; float bias;
memcpy(&bias, (const float *)(op->op_params) + 1, sizeof(float)); memcpy(&bias, (const float *)(op->op_params) + 1, sizeof(float));