vrr/kvcache-ai-ktransformers
2
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2025-09-10 15:29:39 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge 206e51f77e into ee2ede0412

Browse source
This commit is contained in:
Natsu 2025-08-09 13:15:04 +00:00 • committed by GitHub
commit 4f4d7d3696
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 8 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

7
csrc/ktransformers_ext/cpu_backend/shared_mem_buffer.cpp
View file

@ -9,6 +9,7 @@
**/ **/
#include "shared_mem_buffer.h" #include "shared_mem_buffer.h"
#include <cstdio> #include <cstdio>
#include <new>
SharedMemBuffer::SharedMemBuffer() { SharedMemBuffer::SharedMemBuffer() {
buffer_ = nullptr; buffer_ = nullptr;
@ -17,7 +18,7 @@ SharedMemBuffer::SharedMemBuffer() {
SharedMemBuffer::~SharedMemBuffer() { SharedMemBuffer::~SharedMemBuffer() {
if (buffer_) { if (buffer_) {
free(buffer_); ::operator delete(buffer_, std::align_val_t(64));
} }
} }
@ -28,9 +29,9 @@ void SharedMemBuffer::alloc(void* object, std::vector<std::pair<void**, uint64_t
} }
if (size > size_) { if (size > size_) {
if (buffer_) { if (buffer_) {
free(buffer_); ::operator delete(buffer_, std::align_val_t(64));
} }
buffer_ = std::aligned_alloc(64, size); buffer_ = ::operator new(size, std::align_val_t(64));
size_ = size; size_ = size;
for (auto& obj_requests : hist_requests_) { for (auto& obj_requests : hist_requests_) {

5
setup.py
View file

@ -48,7 +48,10 @@ if dev_backend == "xpu":
"pytorch-triton-xpu==3.3.0" "pytorch-triton-xpu==3.3.0"
] ]
else: else:
triton_dep = ["triton>=3.2"] triton_dep = [
"triton >= 3.2; sys_platform != 'win32' and sys_platform != 'Windows'",
"triton-windows >= 3.2; sys_platform == 'win32' or sys_platform == 'Windows'"
]
with_balance = os.environ.get("USE_BALANCE_SERVE", "0") == "1" with_balance = os.environ.get("USE_BALANCE_SERVE", "0") == "1"