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

Apply suggestions from code review

Browse source 
Co-authored-by: Diego Devesa <slarengh@gmail.com>
This commit is contained in:
matt23654 2025-01-03 22:46:35 +00:00 • committed by GitHub
parent c111e8a5b2
commit 4973a298b6
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml/src/ggml-rpc/ggml-rpc.cpp
View file

@ -479,7 +479,7 @@ 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. // 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. // Due to bandwidth constraints, we only call the server init tensor functions if necessary.
// In particular, this is tensors with padding that needs to be cleared, so base tensors only and only misaligned. // In particular, only quantized tensors need padding
if (ggml_is_quantized(tensor->type) && (tensor->ne[0] % 512 != 0) && (tensor->view_src == nullptr)) { if (ggml_is_quantized(tensor->type) && (tensor->ne[0] % 512 != 0) && (tensor->view_src == nullptr)) {
rpc_msg_init_tensor_req request; rpc_msg_init_tensor_req request;
@ -600,7 +600,7 @@ static size_t ggml_backend_rpc_get_max_size(ggml_backend_buffer_type_t buft) {
static size_t ggml_backend_rpc_buffer_type_get_alloc_size(ggml_backend_buffer_type_t buft, const ggml_tensor * tensor) { static size_t ggml_backend_rpc_buffer_type_get_alloc_size(ggml_backend_buffer_type_t buft, const ggml_tensor * tensor) {
// See comments in init_tensor. // See comments in init_tensor.
if (ggml_is_quantized(tensor->type) && (tensor->ne[0] % 512 != 0)) { if (ggml_is_quantized(tensor->type) && (tensor->ne[0] % 512 != 0) && (tensor->view_src == nullptr)) {
ggml_backend_rpc_buffer_type_context * buft_ctx = (ggml_backend_rpc_buffer_type_context *)buft->context; ggml_backend_rpc_buffer_type_context * buft_ctx = (ggml_backend_rpc_buffer_type_context *)buft->context;
auto sock = get_socket(buft_ctx->endpoint); auto sock = get_socket(buft_ctx->endpoint);