We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
There was an error while loading. Please reload this page.
1 parent dc814b8 commit 89faa24Copy full SHA for 89faa24
ggml/src/ggml-cuda/rope.cu
@@ -429,7 +429,7 @@ void ggml_cuda_op_rope_impl(ggml_backend_cuda_context & ctx,
429
GGML_ASSERT( dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
430
// When not fused, src0 and dst types must match
431
// When fused (ROPE+VIEW+SET_ROWS), src0 may be F32 and dst may be F16
432
- GGML_ASSERT(src0->type == dst->type || dst->type == GGML_TYPE_F16);
+ GGML_ASSERT(src0->type == dst->type || (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F16));
433
434
const int64_t ne00 = src0->ne[0]; // head dims
435
const int64_t ne01 = src0->ne[1]; // num heads
0 commit comments