A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 17, 01:45 UTC
[Misc] Apply a couple g++ cleanups (#4719)
Steve Grubb committed
May 10, 2024 at 09:37 UTC
dac6a3f6ed14ea4061b672f9290bfdf8bcdd996d
2 files changed
+1
-2
csrc/cpu/cache.cpp
+1
-1
@@ -84,7 +84,7 @@ void reshape_and_cache_cpu_impl(
84
void copy_blocks(std::vector<torch::Tensor> &key_caches,
85
std::vector<torch::Tensor> &value_caches,
86
const torch::Tensor& block_mapping) {
87
- int num_layers = key_caches.size();
87
+ unsigned num_layers = key_caches.size();
88
TORCH_CHECK(num_layers == value_caches.size());
89
if (num_layers == 0) {
90
return;
csrc/cpu/pos_encoding.cpp
-1
@@ -19,7 +19,6 @@ void rotary_embedding_impl(
19
const int num_tokens) {
20
using scalar_vec_t = vec_op::vec_t<scalar_t>;
21
constexpr int VEC_ELEM_NUM = scalar_vec_t::get_elem_num();
22
- constexpr int ELEM_SIZE = sizeof(scalar_t);
22
23
const int embed_dim = rot_dim / 2;
24
TORCH_CHECK(embed_dim % VEC_ELEM_NUM == 0);