|
@@ -1086,10 +1086,10 @@ bool llama_model_loader::load_all_data(
|
|
|
} else {
|
|
} else {
|
|
|
// If upload_backend is valid load the tensor in chunks to pinned memory and upload the buffers asynchronously to the GPU.
|
|
// If upload_backend is valid load the tensor in chunks to pinned memory and upload the buffers asynchronously to the GPU.
|
|
|
if (upload_backend) {
|
|
if (upload_backend) {
|
|
|
- auto offset = (off_t) weight->offs;
|
|
|
|
|
|
|
+ size_t offset = weight->offs;
|
|
|
alignment = file->read_alignment();
|
|
alignment = file->read_alignment();
|
|
|
- off_t aligned_offset = offset & ~(alignment - 1);
|
|
|
|
|
- off_t offset_from_alignment = offset - aligned_offset;
|
|
|
|
|
|
|
+ size_t aligned_offset = offset & ~(alignment - 1);
|
|
|
|
|
+ size_t offset_from_alignment = offset - aligned_offset;
|
|
|
file->seek(aligned_offset, SEEK_SET);
|
|
file->seek(aligned_offset, SEEK_SET);
|
|
|
|
|
|
|
|
// Calculate aligned read boundaries
|
|
// Calculate aligned read boundaries
|