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

Handle potentially dangerous edge cases.

Browse source
This commit is contained in:
matt23654 2025-01-03 22:14:21 +00:00
parent b66e91b1b2
commit c111e8a5b2
1 changed files with 10 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

11
ggml/src/ggml-rpc/ggml-rpc.cpp
View file

@ -479,7 +479,8 @@ static void ggml_backend_rpc_buffer_init_tensor(ggml_backend_buffer_t buffer, gg
// CUDA backend on the server pads everything to 512 due to CUDA limitations.
// Due to bandwidth constraints, we only call the server init tensor functions if necessary.
if (ggml_is_quantized(tensor->type) && (tensor->ne[0] % 512 != 0)) {
// In particular, this is tensors with padding that needs to be cleared, so base tensors only and only misaligned.
if (ggml_is_quantized(tensor->type) && (tensor->ne[0] % 512 != 0) && (tensor->view_src == nullptr)) {
rpc_msg_init_tensor_req request;
request.tensor = serialize_tensor(tensor);
@ -995,6 +996,14 @@ bool rpc_server::init_tensor(const rpc_msg_init_tensor_req & request) {
GGML_LOG_ERROR("Null buffer for tensor passed to init_tensor function\n");
}
if(tensor->extra != nullptr) {
// This pointer can either be passed around client/server, or probably better stored server-side and kept track of.
// Currently unimplemented.
GGML_LOG_ERROR("tensor->extra populated by the backend, this is currently unsupported.\n");
ggml_free(ctx);
return false;
}
ggml_free(ctx);
return true;
}